JP2914326B2 - Transformation encoding of digital signal enabling reversible transformation - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

[0001] 1. Field of the Invention [0002] The present invention relates to conversion coding of digital signals, and more particularly to a coding method of image signals.

[0002]

2. Description of the Related Art As a reversible encoding method for digital signals based on discrete cosine transform, for example, Japanese Patent Application No. 7-249962 (title of the invention: "Conversion encoding method for digital signals enabling reversible transformation" , Unpublished at the time of filing of the present application), and also by the present inventor on August 23, 1996.
There is a technology described in a prior application such as a dated patent application (title of the invention: "transformation encoding method of digital signal enabling reversible transformation").

[0003] These are modifications of the discrete cosine transform to enable a reversible transform. These schemes have two features, one of which is to approximate a discrete cosine transform by a linear transform by an integer matrix, and the other is to remove redundancy included between transform coefficients by reversible quantization. That is.

[0004] The principle of the 8-ary reversible discrete cosine transform according to this method will be described below. The original 8-ary discrete cosine transform converts the original signal vector (x ₀ , x ₁ ,..., X ₇ ) into a conversion coefficient (X ₀ , X ₁ ,..., X ₇ ) according to the following equation (1). It is.

[0005]

(Equation 2)

Here, c ₁ ,..., C ₇ are given by the following equation (2).

[0007]

(Equation 3)

In the 8-ary reversible discrete cosine transform described in the patent application, the 8-ary discrete cosine transform represented by the above equation (1) is approximated by the following equation (3).

[0009]

(Equation 4)

Here, a ₁ ,..., A ₇ are natural numbers. A ₁ ,..., A ₇ are obtained by multiplying each row vector of the discrete cosine transform by a constant and rounding it to an integer.

(X ₀ ,..., X ₇ ) obtained by the above equation (3) is
Except for being multiplied by a constant, the value is close to the original discrete cosine transform coefficient. Since (X ₀ ,..., X ₇ ) are integer values, lossless encoding is possible.

However, the conversion coefficient X obtained by the above equation (3)
_0, ..., X ₇ are not independent of each other, as a vector signal having redundancy. Assuming that the absolute value of the determinant of the transformation matrix is D, the density of points that can be taken in the transformation domain is 1 / D. That is, in the conversion domain, the points that can be taken out of all the integer vectors are 1 / D of the whole, and the remaining 1-1 / Ds are useless points that cannot be actually taken. Making this useless point a target of encoding causes a reduction in the efficiency of compression encoding.

Therefore, in order to remove this redundancy, lossless encoding is performed. However, since it is difficult to directly define reversible quantization in an eight-dimensional space, the above equation (3) is decomposed according to a fast algorithm. Then, lossless quantization is performed on the conversion results of the obtained individual partial conversions.

The conversion of the above equation (3) can be decomposed as follows.

[0015]

(Equation 5)

However, in the above equation (10), a ₄ = 1. This matrix of the equation (10) includes only a a _4, because even there is no problem as a ₄ = 1.

In this way, it can be decomposed into eight 2 × 2 matrix transformations and one 4 × 4 matrix transformation.

Redundancy occurs in the individual transforms from equations (4) to (12). For example, in the case of the conversion of the above equation (4), since the absolute value of the determinant of the conversion matrix is 2, the integer vector that can be actually obtained as a conversion result is
Only 1/2 of the whole. Such redundancy is removed for each individual transform.

For this purpose, as shown in FIG. 15, the conversion result is reversibly quantized for each conversion (see reversible quantization Q arranged at the subsequent stage of each converter in FIG. 15).

In the reversible quantization, a transform point obtained by transforming an integer lattice point is quantized using the periodicity of the structure. In the case of the conversion using the 2 × 2 matrix among the conversions of Expressions (4) to (12), the absolute value of the determinant of the conversion matrix is D, and the coordinates of the conversion point obtained by converting the integer vector are (Y ₁ , Y ₂ ), the entire structure of the conversion point (Y ₁ , Y ₂ ) has a period D in each axis direction. Therefore, the area defined by the following equation (13):

[0021]

(Equation 6)

(Where N ₁ and N ₂ are multiples of D, and these are hereinafter referred to as “quantization periods”), and a quantization value of a transformation point within this region is defined as It is artificially defined in advance. The correspondence between the defined transformation points and the quantized values is shown in a table (hereinafter, this table is referred to as a "quantization correspondence table").
), And as described below,
Quantization is performed using this table.

An actual quantization procedure is as follows.
_First , quotients b ₁ , b ₂ and remainders r ₁ , r ₂ when the coordinates Y ₁ , Y ₂ of the transformation point are divided by the quantization periods N ₁ , N ₂ are obtained. Then, the remainder (r ₁ , r ₂ ) is calculated by the above equation (1)
It is included in the basic area shown in 3).

Next, determine the using a quantization correspondence table defined above (r _1, r ₂₎ quantized values (q _1, q ₂₎ (hereinafter referred to as local quantization). Then, the quotients b ₁ and b ₂ and the local quantization value q
_1, from the _{q 2 (Y 1, Y 2} ) quantization values of the (Y _q1, Y _q2),
It is calculated by equation (14).

[0025]

(Equation 7)

Here, M ₁ and M ₂ are natural numbers, and q
₁ and q ₂ represent the dynamic range. These M ₁ and M ₂ are
The scale of the quantized value represents the size of the basic region when it is scaled, and represents the period of inverse quantization described below (for this reason, hereinafter, M ₁ and M ₂ are referred to as “inverse quantization period”). I will call it).

Next, the inverse quantization of the 2 × 2 matrix transformation will be described. Inverse quantization can be performed in the same procedure as quantization.

Firstly, divided by the quantization value Y _q1, inverse quantization period described above Y _q2 respectively M _1, M _2, obtains the quotient b _'1, b' ₂ and a remainder q _'1, q' _2.

Next, with reference to a table describing a correspondence relationship opposite to the quantization correspondence table (hereinafter referred to as an “inverse quantization correspondence table”), (q ′ ₁ , q ′ ₂ ) is corresponded ( r ′ ₁ , r ′ ₂ ). Then, the inverse quantization value (Y ₁ , Y ₂ ) is calculated by the following equation (15).

[0030]

(Equation 8)

The above is the quantization and inverse quantization of the 2 × 2 matrix conversion.

In the quantization of (X ₁ , X ₇ , X ₃ , X ₅ ) obtained by the conversion of the 4 × 4 matrix shown in the above equation (12), the following relationship is used. That is, when (X ₁ , X ₇ ) is determined, a certain integer vector (s ₃ , s ₅ ) is determined, and the possible values of (X ₃ , X ₅ ) are limited to the following equation (16). Use.

[0033]

(Equation 9)

Here, g and h are a ₁ ,
From a ₃ , a ₅ , and a ₇ , it is an integer represented by the following equation (17).

[0035]

(Equation 10)

As described above, when a certain (X ₁ , X ₇ ) is determined, the structure of the entire possible value of the conversion coefficient (X ₃ , X ₅ ) is represented by the vector (h, g), (-g, h ) Is a structure shifted by the integer vector (s ₃ , s ₅ ) (hereinafter referred to as “representative element”). Utilizing such a relationship, (X ₁ , X ₇ ) and (X ₃ , X ₅ ) are individually quantized.

For the conversion coefficients X ₁ and X ₇ , the widths k ₁ and k ₇
( _Where k ₁ and k ₇ are natural numbers) are linearly quantized to obtain quantized values X _q1 and X _q7 . At this time, the quantization residual (r
₁ , r ₇ ) is also determined. On the other hand, the transform coefficients (X ₃ , X ₅ ) are quantized using the relationship of the above equation (16). In this quantization, (X ₃ , X ₅ ) is represented by (p ₃ , p ₅ ) = (X ₃ −s ₃ , X ₅ −s ₅ ) representing a global lattice structure (hereinafter referred to as a global signal). (S ₃ , s ₅ ) representing parts and local signal components
(Hereinafter referred to as “representative element”) and quantized.

The quantization of the global signal (p ₃ , p ₅ ) is performed by the method disclosed in Japanese Patent Application No. Hei 7-249962 and the method disclosed in
This is different from the three-day patent application method.

In the Japanese Patent Application No. 7-249962, quantization is performed in the same manner as the above-mentioned reversible quantization of 2 × 2 matrix transformation. In the above-mentioned patent application filed on Aug. 23, 1996, linear quantization is applied to the global quantization value. (P _q3 , p _q5 ). Among them, the above-mentioned patent application filed on Aug. 23, 1996 can reduce the difference from the original discrete cosine transform coefficient.

On the other hand, the quantization of the representative elements (s ₃ , s ₅ ) is performed using a table defining the quantization in any of the above-mentioned prior patent applications. However, (s _3, s ₅₎ instead of obtaining the quantized value (s _q3, s _q5) directly from the quantized residual of the aforementioned (r _1, r _7), and (r _1, r ₇₎ (s _q3, s _q5) quantization value using a table describing the correspondence between (s _q3, s _q5)
Ask for. The global quantization values (p _q3 , p _q5 ) and the local quantization values (s _q3 , s _q5 ) thus obtained are combined to calculate the quantization values (X _q3 , X _q5 ).

The inverse quantization of the 4 × 4 matrix conversion is performed as follows. In the inverse quantization of the quantization values (X _q1 , X _q7 ), first, linear inverse quantization is performed.

Next, the transform coefficients (X ₁ , X ₇ ) are calculated by adding the quantization residuals (r ₁ , r ₇ ) to this. The quantization residuals (r ₁ , r ₇ ) are obtained from a local quantization value (s _q3 , s _q5 ) obtained by inverse quantization of (X _q3 , X _q5 ) using a table.

On the other hand, in the inverse quantization of the quantization values (X _q3 , X _q5 ), first, they are _converted into global quantization values (p _q3 , p _q5 ) and local quantization values (s _q3 , s _q5 ). To separate.

Next, from the global quantization values (p _q3 , p _q5 ),
In the above-mentioned Japanese Patent Application No. 7-249962, a table is used, and in the above-mentioned patent application filed on August 23, 1996, the global signal (p ₃ , p ₅ ) is obtained by using the inverse transform of the above equation (16).

On the other hand, from the local quantization values (s _q3 , s _q5 ), the quantization residuals (r ₁ , r ₇ ) are obtained using a table describing the inverse relationship to that used in the quantization. Ask for. From the quantization residual (r ₁ , r ₇ ), (X ₁ , X ₇ ) is obtained as described above, and the representative element (s ₃ , s ₅ ) is obtained by referring to the table from (X ₁ , X ₇ ).
Is found. Finally, the transform coefficients (X ₃ , X ₅ ) are obtained by adding the (s ₃ , s ₅ ) to the global signal (p ₃ , p ₅ ).
Thus, the inverse quantization is completed.

The lossless quantization of the 2 × 2 or 4 × 4 matrix conversion described above is performed after each conversion from Expressions (4) to (12) as shown in FIG. It is possible to eliminate the included redundancy.

As a result, lossless encoding can be realized without deteriorating the encoding efficiency. Note that the quantization period N _i ,
By appropriately adjusting the value of the inverse quantization cycle M _i , it is possible to make the finally obtained transform coefficient of the eight-way invertible discrete cosine transform close to that of the original discrete cosine transform.

[0048]

However, the 8-ary reversible discrete cosine transform described in the above-mentioned prior application has a problem that compatibility with the conventional 8-ary discrete cosine transform is not sufficient. That is, when the signal coded using the 8-ary reversible discrete cosine transform described above is decoded using the inverse 8-ary discrete cosine transform of the conventional method, or the quality of the reproduced signal obtained in the reverse case is sufficient. Was not.

The reason is that 8
This is because there is a difference between the dynamic coefficient of the transform coefficient of the original reversible discrete cosine transform and the transform coefficient of the original 8-ary discrete cosine transform.

As described above, in the 8-ary reversible discrete cosine transform according to the prior application, the transform coefficients obtained by properly matching the parameters such as the quantization cycle and the inverse quantization cycle can be used as the original 8-ary discrete cosine transform coefficients. Was approaching. However, there is a limit to this, and there is a problem that the dynamic range of the obtained transform coefficients slightly deviates from the original 8-ary discrete cosine transform coefficients.

Accordingly, the present invention has been made in view of the above circumstances, and an object of the present invention is to make the transform coefficient of the reversible discrete cosine transform close to the original discrete cosine transform coefficient, thereby achieving the discrete cosine transform of the conventional method. It is an object of the present invention to provide an image coding method that enhances compatibility between a transform and a reversible discrete cosine transform. More specifically, an object of the present invention is to improve the quality of a reproduced signal obtained when a signal encoded using an 8-ary reversible discrete cosine transform is decoded using a conventional 8-ary discrete cosine transform. It is another object of the present invention to provide an image coding method as described above.

[0052]

In order to achieve the above object, the first invention of the present application is a method for encoding an image signal, and performs a reversible discrete cosine transform on the image signal to obtain a transform coefficient. Means for obtaining, a means for storing a quantization table for JPEG compatibility, and means for performing variable length coding on the transform coefficients and the quantization table for JPEG compatibility and outputting a coded signal. .

The conventional image coding method using the reversible discrete cosine transform does not use a quantization table. In the first invention of this application, by using this quantization table, JPEG (Joint Photographic Coding), which is an international standard for image coding based on discrete cosine transform, is used.
Experts Group).

The first invention of the present application has an advantage that it can be easily realized only by preparing a storage device for storing a fixed quantization table.

The second invention of the present application is a method for encoding a moving image signal, comprising means for estimating a motion between a previously encoded image and an image to be encoded to obtain a motion vector. Means for performing motion compensation using a motion vector to generate a predicted image from a previously coded image, means for subtracting the predicted image from the image to be coded, and generating a predicted error image, Means for performing a reversible discrete cosine transform on, and calculating a transform coefficient;
(Motion Picture Experts Group) means for storing a quantization matrix for compatibility, variable length coding of a transform coefficient, a quantization matrix for MPEG compatibility and a motion vector,
Means for outputting the result.

Similarly to the first invention of the present application, the second invention of the present application improves the compatibility with MPEG, which is an international standard for video coding, by using a quantization matrix that has not been used conventionally. I have. The second invention of the present application also has an advantage that it can be easily realized simply by preparing a storage device for storing a fixed quantization matrix value, similarly to the first invention of the present application.

The third invention of the present application is a system for encoding an image signal, which performs a reversible discrete cosine transform on the image signal to obtain a reversible discrete cosine transform coefficient.
Means for performing a discrete cosine transform on an image signal to obtain a discrete cosine transform coefficient, and calculating a JPEG compatible quantization table that reduces the difference between the inverse quantized value of the reversible discrete cosine transform coefficient and the discrete cosine transform coefficient And a means for outputting a lossless discrete cosine transform coefficient and a signal encoded by performing variable length encoding on a quantization table for JPEG compatibility.

In the third aspect of the present invention, the compatibility with JPEG can be adaptively improved by sequentially calculating an appropriate quantization table value, as compared with the case where a fixed quantization table is used.

The fourth invention of the present application is a system for encoding a moving image signal, which is a means for estimating a motion between a previously encoded image and an image to be encoded to obtain a motion vector. Means for performing motion compensation using a motion vector to generate a first predicted image from an image encoded in the past, and subtracting the first predicted image from the image to be encoded to obtain a first predicted image. Means for creating an error image; means for performing a reversible discrete cosine transform on the first prediction error image to calculate a reversible discrete cosine transform coefficient; Means for generating a second prediction error image, performing discrete cosine transform on the second prediction error signal to calculate a discrete cosine transform coefficient, and inverse quantization of the reversible discrete cosine transform coefficient. Value and
MPEG that reduces the difference between discrete cosine transform coefficients
Calculates and outputs the compatible quantization matrix,
Means for outputting an inverse quantized value of a reversible discrete cosine transform coefficient when using a quantization matrix for MPEG compatibility, and inverse discrete cosine transform for the inverse quantized value to calculate a prediction error reproduced image Means and the second
And a means for adding a predicted image to generate a reproduced image, and a means for performing motion compensation using a motion vector to generate a second predicted image from the reproduced image.

In the fourth invention of the present application, as in the third invention of the present application, compared with the case of using a fixed quantization matrix,
Compatibility with MPEG can be adaptively improved.

A fifth aspect of the present invention is a preferred embodiment of the image encoding system of the third aspect of the present invention or the fourth aspect of the present invention, and calculates the JPEG / MPEG compatible quantization table / matrix. Means for inversely quantizing the reversible discrete cosine transform coefficients based on the values of a quantization table / matrix for JPEG / MPEG compatibility,
Means for determining an inverse quantization value, discrete cosine transform coefficients,
Means for calculating the mean square of the difference from the inverse quantization value for each coefficient, and means for correcting and outputting the value of the JPEG / MPEG compatible quantization table / matrix based on the mean square It is characterized by.

According to the fifth aspect of the present invention, it is possible to sequentially calculate a quantization table / matrix that totally minimizes the difference between the reversible discrete cosine transform coefficient and the original discrete cosine transform coefficient. The quantization table / matrix obtained by this method is
This is optimal in that the decoding image quality is best when this is used.

Further, the sixth invention of the present application is an embodiment of the image coding system of the third or fourth invention, wherein the means for calculating the JPEG / MPEG compatible quantization table / matrix includes: From the reversible discrete cosine transform coefficient X _q (i, j), a value equivalent to a value obtained by dividing a value before being converted to an integer by inverse quantization by a quantization table / matrix is obtained, and this is calculated as X ′ (i, j) And a value obtained by adding or subtracting a minute number to or from the discrete cosine transform coefficient X _T (i, j) according to the value of the reversible discrete cosine transform coefficient X _q (i, j) is represented by X ′ _T (i, j). ), X ′ (i, j) and X ′
Means for calculating and outputting a quantization table / matrix by an operation based on a second moment related to _T (i, j), and inversely quantizing the reversible discrete cosine transform coefficient using the quantization table / matrix, Means for outputting a quantized value.

In the sixth aspect of the present invention, the processing amount required for the quantization table / matrix calculation can be significantly reduced by introducing the approximate calculation.

The seventh invention of the present application is an aspect of the image coding system according to the third or fourth invention, wherein the means for calculating the JPEG / MPEG compatible quantization table / matrix includes: JPEG / MP of invention 5
It is characterized by having means for calculating a quantization table / matrix for EG compatibility and means for calculating a quantization table / matrix for JPEG / MPEG compatibility of the sixth invention.

In the seventh invention of the present application, by appropriately combining the fifth invention and the sixth invention, it is possible to suppress the processing amount and calculate a quantization table / matrix very close to the optimum one. I do.

The eighth invention of the present application is a conversion method for performing a reversible 8-ary discrete cosine transform in the vertical and horizontal directions on a two-dimensional signal of an 8 × 8 block to obtain a transform coefficient. The matrix through which the components are calculated

[0068]

[Equation 11]

The DC component of the reversible discrete cosine transform and the original discrete cosine transform are changed by changing the quantization correspondence table used for any three of the six reversible quantizations based on the above. It is characterized by eliminating the dynamic range difference between the DC component.

In the eighth invention of the present application, since the dynamic range difference between the 8 × 8 reversible discrete cosine transform coefficient and the DC component of the original 8 × 8 discrete cosine transform coefficient can be completely compensated for, the reversible discrete cosine transform is used. The encoded image is
The image quality when decoding is performed using the inverse discrete cosine transform can be greatly improved.

[0071]

Embodiments of the present invention will be described below with reference to the drawings.

[0072]

Embodiment 1 FIG. 1 is a block diagram showing a configuration of an embodiment of the first invention of the present application. Referring to FIG. 1, a reversible discrete cosine transform circuit 1 performs reversible discrete cosine transform on an image for each block and outputs transform coefficients. The variable length encoder 2 outputs the output of the reversible discrete cosine transform circuit 1 and J
The variable-length encoding is performed on the quantization table for PEG compatibility and the result is output.

Next, a method of calculating the value of the quantization table for JPEG compatibility will be described.

The quantization table is a table used to change the quantization width for each discrete cosine transform coefficient.
It is adopted by JPEG, which is an international standard for still image coding. Each value in this table determines the quantization width of the transform coefficient at that position. That is, i of the quantization table
The value of row j column (i = 0,..., 7; j = 0,.
j), in JPEG (extended system), the (i, j) component of the transform coefficient (where (i, j) ≠
(0,0)) and the inverse quantized value X _q (i, j)
The relationship of (i, j) is as shown in the following equations (18) and (19).

[0075]

(Equation 12)

Here, s is a parameter called a quantization scale, and i and j do not depend. Also, Sign
(Y) takes a value represented by the following equation (20).

[0077]

(Equation 13)

As can be seen from the above equations (18) and (19), the values of the transform coefficients other than X _q (0,0) are multiplied by about W (i, j) s / 16 by this inverse quantization. You.

As described above, in the decoder, the quantized transform coefficient value X _q (i, j) is scaled by inverse quantization before performing the inverse discrete cosine transform. Then, the magnification can be freely set for each transform coefficient by using a quantization table. Therefore, the transform coefficient of the reversible discrete cosine transform is
When decoding by the conventional inverse discrete cosine transform, the quantization table value W
If the dynamic range of the transform coefficient of the reversible discrete cosine transform is corrected using (i, j), the decoded image quality can be improved.

Next, after describing the difference in the dynamic range between the transform coefficient of the reversible discrete cosine transform and the original discrete cosine transform coefficient, a method of calculating the value of the quantization table will be specifically described.

As described above, there is a difference in the dynamic range between the transform coefficient of the reversible discrete cosine transform and the original discrete cosine transform coefficient. This, as described below,
This occurs because the ratio of expansion by linear transformation using an integer matrix does not completely match the ratio of reduction by lossless quantization using a table.

In the reversible discrete cosine transform, when the original discrete cosine transform is approximated by an integer matrix linear transform, the transform base is enlarged, so that the obtained result is also enlarged.

On the contrary, in the reversible quantization, the value after quantization has a smaller dynamic range than that before quantization.

If the expansion is just canceled by the compression, there is no dynamic range difference. However, in general, the ratio of the expansion and the reduction does not completely match. The coefficients have a dynamic range difference from the original discrete cosine transform coefficients. This difference occurs in each of the 2 × 2 and 4 × 4 partial transforms shown in FIG.

The original 8-ary discrete cosine transform can be decomposed as shown in FIG. 16, but the difference between the partial transforms is specifically determined by the transform obtained by the corresponding converter in FIG. 15 and FIG. The resulting dynamic range difference, ie, converter 1
60 and converter 180, converter 161 and converter 181, converter 162 and converter 182, converter 163 and converter 18
3, converter 164 and converter 184, converter 165 and converter 185, converter 166 and converter 186, converter 167
And the dynamic range difference between the conversion results obtained by the converter 187 and the converter 168 and the converter 188.

The difference in the dynamic range of the finally obtained transform coefficient is obtained by superimposing the difference caused by the partial transform that is passed on the way.

Next, a method of calculating the difference between the dynamic ranges and determining the quantization matrix value based on the difference will be described.

First, in the case of the 2 × 2 matrix conversion, the dynamic range of the reversible discrete cosine transform is obtained. The coordinates of the conversion points obtained by converting the integer vector using the 2 × 2 matrix are (Y ₁ , Y ₂ ), and the size of the row vector of the conversion matrix is L. At this time, the values of Y ₁ and Y ₂ are
It becomes L times that of the original discrete cosine transform. This is because, in each partial transform of the original discrete cosine transform shown in FIG. 16, the magnitude of the vector does not change before and after the transform, whereas in the case of the reversible discrete cosine transform, the transform result (Y
_This is because the magnitude of ( ₁ , Y ₂ ) is L times the magnitude of the vector before conversion.

On the other hand, in the reversible quantization after the linear transformation, the quantization cycle is (N ₁ , N ₂ ) and the inverse quantization cycle is (M ₁ ,
M ₂ ), the values of Y ₁ and Y ₂ are respectively about M ₁ / N ₁
Approximately M ₂ / N ₂ times. Therefore, the quantized values of Y ₁ and Y ₂ obtained after the reversible quantization are about M ₁ L / N ₁ times and about M ₂ L / N ₂ times as compared with the case of the original discrete cosine transform.

Next, the case of the 4 × 4 matrix conversion shown in the above equation (12) will be described.

In this case, assuming that the magnitude of the row vector is L, (X ₁ , X ₇ , X ₃ , X
₅ ) is about L times larger than the original discrete cosine transform.

On the other hand, in the reversible quantization,
In the method described in the patent application filed on Aug. 23, 2008, the transform coefficients (X ₁ , X ₇ ) are linearly quantized with widths k ₁ and k ₇ , respectively.
As for the transform coefficients (X ₃ , X ₅ ), the global signal portion is linearly quantized with widths L ₃ and L ₅ .

Therefore, in this quantization, X ₁ and X ₇ are respectively 1 / k ₁ times and 1 / k ₇ times, and X ₃ and X ₅ are respectively 1 / k.
L _3-fold, it is compressed to 1 / L _5-fold. For this reason, the quantized values of (X ₁ , X ₇ , X ₃ , X ₅ ) obtained after the quantization are approximately L / k ₁ times, respectively, as compared with the case of the original discrete cosine transform.
L / k is ₇ times, L / L is ₃ times, and L / L is ₅ times. 4x
(4) When the method described in Japanese Patent Application No. 7-249962 is used as the reversible quantization, the matrix of the following equation (21) is used.

[0094]

[Equation 14]

The quantization cycle and the inverse quantization cycle of the reversible quantization according to (N ₁ ⁽²⁾ , N ₂ ⁽²⁾ ), (M ₁ ⁽²⁾ , M ₂ ⁽²⁾ ), respectively.
Then, by using the following equation (22), the above holds true.

[0096]

(Equation 15)

Here, m ₃ and m ₅ are the dynamic ranges of the local quantization values of X ₃ and X ₅ .

The dynamic range of the finally obtained conversion coefficient is calculated based on the dynamic range of the conversion result of the 2 × 2, 4 × 4 conversion calculated as described above. Here, in the conversion using the same matrix, the dynamic range is calculated assuming that quantization is performed using the same quantization correspondence table.

The matrix of FIG. 15 given by the following equations (23) and (24)

[0100]

(Equation 16)

The quantization periods in the reversible quantization for (N ₁ ⁽⁰⁾ , N ₂ ⁽⁰⁾ ) and (N ₁ ⁽¹⁾ , N ₂ ⁽¹⁾ ) are as follows:
The inverse quantization periods are (M ₁ ⁽⁰⁾ , M ₂ ⁽⁰⁾ ), (M ₁ ⁽¹⁾ ,
M ₂ ⁽¹⁾ ).

At this time, the dynamic range of the transform coefficient X _qi (i = 0,..., 7) of the 8-ary reversible discrete cosine transform is
If β (i) is assumed to be “1” in the original discrete cosine transform, the value of β (i) is expressed by the following equation (25).
To (32).

[0103]

[Equation 17]

Here, L ⁽¹⁾ and L are respectively expressed by the following equations (3)
3) and (34), and represent the magnitudes of the row vectors of the transformation matrices of the above equations (24) and (12), respectively.

[0105]

(Equation 18)

Further, in the case of the transform coefficient of the 8 × 8 reversible discrete cosine transform, the (i, j) component X _q (i,
Assuming that the dynamic range of j) is β (i, j), when the dynamic range of the original 8 × 8 discrete cosine transform is “1”, β (i, j) in Expressions (25) to (32) is obtained. ) Can be expressed as the following equation (35).

[0107]

[Equation 19]

If the values of β (i) are different from each other in the horizontal direction and the vertical direction, β (i) in the horizontal and vertical directions
As β _h (i) and β _v (i), respectively,
j) is represented by the following equation (36).

[0109]

(Equation 20)

Next, a method of determining the quantization table value W (i, j) from the dynamic range β (i, j) of each transform coefficient thus obtained will be described.

As described above, the dynamic range of the quantized value X _q (i, j) is set to W
(I, j) s / 16 times. Therefore, in order to correct the dynamic range, the quantization table value W (i, j) may be determined so as to satisfy the following equation (37).

[0112]

(Equation 21)

Considering that the quantization table value W (i, j) is an integer, W (i, j) is determined by the following equation (38).

[0114]

(Equation 22)

Here, rnd (x) is a function that returns a value obtained by rounding off the decimal part of x. If the above equation (38)
Is outside the range of possible values of the quantization table, the integer within the range closest to the value is set as the quantization table value.

[0116]

Embodiment 2 FIG. 2 is a block diagram showing a configuration of an embodiment of the second invention of the present application. Here, the reversible discrete cosine transform circuit 1 is the same as in the embodiment shown in FIG.

Referring to FIG. 2, motion estimation circuit 5 has a storage device (not shown) for storing previously coded images, and stores the image stored in this storage device and the current code. Motion estimation is performed between images to be converted, and a motion vector is output.

The predictor 4 performs motion compensation on a reference image for motion estimation based on the motion vector obtained by the motion estimating circuit 5 to create a predicted image. Then, the prediction image is output.

The adder 3 subtracts the predicted image output from the predictor 4 from the input image to be encoded, and creates a prediction error image.

The variable-length encoder 6 performs variable-length encoding on the transform coefficient output from the reversible discrete cosine transform circuit 1, the motion vector output from the motion estimating circuit 1, and the quantization matrix for MPEG compatibility. Output the result.

Next, the operation of the circuit of this embodiment shown in FIG. 2 will be described.

First, in the motion estimation circuit 5, motion estimation is performed between the image to be coded and the already coded image, and a motion vector is calculated for each macroblock.
Then, the predictor 4 generates a predicted image from the already encoded image based on the obtained motion vector. This predicted image is subtracted from the encoding target image in the adder 3 to obtain a prediction error image. However, in this prediction error image, the original signal value is entered for the intra block instead of the prediction error.

This predictive error image is subjected to a reversible discrete cosine transform for each block in a reversible discrete cosine transform circuit 1 to calculate a transform coefficient. This conversion factor,
The motion vector and a predetermined quantization matrix for MPEG compatibility are subjected to variable-length encoding in a variable-length encoder 6.

Next, the quantization matrix value will be described.

The quantization matrix is a matrix used to change the quantization width for each discrete cosine transform coefficient, and is employed in MPEG which is an international standard for moving picture coding. Basically, it is the same as the quantization table used in JPEG, and each component of this matrix determines the quantization width of the transform coefficient at that position.

The (i, j) component of the quantization matrix is represented by W
When expressed by (i, j) (i = 0,..., 7; j = 0,..., 7), for the transform coefficients other than the intra-block DC component, the quantization value of the (i, j) component X _q (i, j)
And the inverse quantization value X (i, j) are expressed by the following equation (39):
This is as shown in (40).

[0127]

(Equation 23)

Here, s is a parameter called a quantization scale, and does not depend on i and j. Also, Sign
(Y) is the value represented by the above equation (20), and k is the following equation (4)
This is the value represented by 1).

[0129]

(Equation 24)

As can be seen from the above equations (39) and (40), the value of the transform coefficient is about W (i,
j) Multiplied by s / 16. Focusing on this point, the same method as that for obtaining the quantization table for JPEG compatibility is used.
Obtain a quantization matrix for EG compatibility. That is, the quantization matrix value W (i, j) is determined according to the above equation (38) from the dynamic range β (i, j) of each transform coefficient obtained by the above equation (35).

[0131]

Third Embodiment FIG. 3 is a block diagram showing a configuration of an embodiment of the third invention of the present application. Here, the reversible discrete cosine transform circuit 1 and the variable length encoder 2 are shown in FIG.
Is the same as that of the first embodiment shown in FIG.

Referring to FIG. 3, discrete cosine transform circuit 1
1 performs discrete cosine transform of the image for each block and outputs a transform coefficient. The JPEG / MPEG compatible quantization table / matrix generation device 12 generates a quantization table / matrix based on the transform coefficient values output from the reversible discrete cosine transform circuit 1 and the discrete cosine transform circuit 11, respectively, and Output. This JPEG / MP
As the EG compatible quantization table / matrix generation device 12, for example, the fifth, sixth, and seventh inventions described above can be used.

Next, the operation of the circuit of the embodiment shown in FIG. 3 will be described.

The process until the image signal is subjected to the reversible discrete cosine transform by the reversible discrete cosine transform circuit 1 is the same as that of the embodiment shown in FIG. Referring to FIG. 3, in this embodiment, at the same time, discrete cosine transform circuit 11 performs discrete cosine transform on the image signal for each block.

Then, a quantization table / matrix value is calculated in the JPEG / MPEG compatible quantization table / matrix generation device 12 from the values of the reversible discrete cosine transform coefficient and the discrete cosine transform coefficient.

The quantization table / matrix value is determined so as to reduce the difference between the inverse quantized value of the reversible discrete cosine transform coefficient obtained on the decoder side and the original discrete cosine transform coefficient. More specifically, in a period (for example, a picture) for calculating the quantization table / matrix, a quantization table / matrix value that reduces the difference between the average values of the two is assigned to each (i, j) component. calculate. Then, in the variable length encoder 2, the lossless discrete cosine transform coefficient, the motion vector, and the quantization table / matrix value are subjected to variable length encoding.

[0137]

Fourth Embodiment FIG. 4 is a block diagram showing a configuration of an embodiment of the fourth invention of the present application. Referring to FIG.
The reversible discrete cosine transform circuit 1, the adder 3, the predictor 4, the motion estimating circuit 5, and the variable length encoder 6 are the same as those in the embodiment shown in FIG. Also,
The discrete cosine transform circuit 11 is the same as that of the third embodiment shown in FIG.

Referring to FIG. 4, a JPEG / MPEG compatible quantization table / matrix generation device 25 is basically the same as that shown in FIG. 3, except that only the quantization table / matrix values are used. Instead, the inverse quantized value obtained when the reversible discrete cosine transform coefficient is inversely quantized using this value is output at the same time.

The inverse discrete cosine transform circuit 21 performs an inverse discrete cosine transform on a block-by-block basis, and outputs a prediction error reproduced image as a result.

The first adder 24 converts the prediction error reproduced image into
A reproduced image is created in addition to a second predicted image described later.

The predictor 22 holds the reproduced image output from the first adder 24, and, based on the motion vector obtained by the motion estimating circuit 5, from the already decoded image.
A second predicted image is created.

The second adder 23 creates a second prediction error image by subtracting the second prediction image from the image to be encoded.

Next, the operation of the circuit of the embodiment shown in FIG. 4 will be described.

A motion estimation circuit 5 obtains a motion vector, a predictor 4 prepares a predicted image, an adder 3 prepares a prediction error image, and the reversible discrete cosine transform circuit 1 performs a reversible discrete cosine transform. Up to this point, the configuration is the same as that of the second embodiment shown in FIG.

The quantization matrix value obtained by the JPEG / MPEG compatible quantization table / matrix generation device 25 is basically the same as that of the embodiment described with reference to FIG. However, in the case of this embodiment shown in FIG. 4, not only the quantization matrix but also the inverse quantization value when the quantization matrix value is used is obtained at the same time.

The inversely quantized value is subjected to the inverse discrete cosine transform in the inverse discrete cosine transform circuit 21. As a result, when decoding is performed using the ordinary inverse discrete cosine transform, a prediction error reproduced image which is assumed to be obtained on the decoding side is obtained. Then, in the first adder 24, a second predicted image is added to the predicted error reproduced image to generate a reproduced image. This reproduced image corresponds to a reproduced image that is assumed to be obtained on the decoding side when decoding is performed using a normal discrete cosine inverse transform.

[0147] The predictor 22 performs motion estimation using the motion vector obtained by the motion estimation circuit 5, thereby generating a second predicted image from the previously decoded reproduced image.
The second prediction image is subtracted from the encoding target image in the second adder 23 to obtain a second prediction error image.

This second prediction error image is subjected to discrete cosine transform in the discrete cosine transform circuit 11. Thereby, a discrete cosine transform coefficient of the prediction error image to be obtained on the decoder side is obtained. Therefore, the quantization table / matrix may be determined so that the value obtained by inversely quantizing the reversible discrete cosine transform coefficient is close to the value.

Therefore, this discrete cosine transform coefficient is expressed as J
Used as input to the quantization table / matrix generation device for PEG / MPEG compatibility.

[0150]

[Fifth Embodiment] FIG. 5 is a diagram for explaining an embodiment of the fifth invention of the present application, and is a block diagram showing an example of the configuration of a quantization table / matrix generation device for JPEG / MPEG compatibility. is there.

With reference to FIG. 5, the inverse quantizer 30
According to the inverse quantization method of PEG or JPEG, the reversible discrete cosine transform coefficient is inversely quantized and output. The MSE calculation circuit 31 for each coefficient calculates the square mean of the difference between the discrete cosine transform coefficient and the inverse quantized value output from the inverse quantizer 30 for each transform coefficient, and outputs the result.

Quantization table / matrix correction device 32
Gives the quantization table / matrix values to the inverse quantizer 30 and the resulting MSE for each coefficient
(MeanSquare Error) Check the output of the calculation circuit 31.

Then, a quantization matrix value whose value is minimized is obtained. This is performed for each transform coefficient.

Finally, the MSE calculation circuit 3 for each coefficient
The value of the quantization table / matrix that minimizes the output of 1 is output.

Next, the operation of the circuit of the embodiment shown in FIG. 5 will be described.

The transform coefficients (i,
j) component is X _q (i, j), and its dequantized value is X (i, j).
j), and the (i, j) component of the discrete cosine transform coefficient is represented by X _T (i, j). The reversible discrete cosine transform coefficient X _q (i, j) is inversely quantized by the inverse quantizer 30 using the quantization table / matrix value W (i, j) input from the quantization table / matrix correction device 32. Be transformed into This inverse quantization is performed according to the above equations (18) and (19) in the case of JPEG, and is performed according to the above equations (39) and (40) in the case of MPEG. As a result, an inverse quantization value X (i, j) is obtained.

In the MSE calculating circuit for each coefficient, the inverse quantized value X (i, j) and the discrete cosine transform coefficient X _T (i, j)
Then, the following equation (42) is calculated, and for each (i, j),
The mean square value mse (i, j) of e (i, j) is calculated and output.

[0158]

(Equation 25)

This average is obtained for the same (i, j) for blocks included in a cycle (for example, a picture) for calculating the quantization matrix. mse
(I, j) is the quantization table / matrix correction device 3
2 is input.

Quantization table / matrix correction device 32
Holds the value of W (i, j) that minimizes this mse (i, j) and its minimum value. Then, a certain W (i,
If the value of mse (i, j) obtained for the value of j) is smaller than the already obtained minimum value of mse (i, j), the minimum value is updated. Give W
Holds the value of (i, j).

This is performed on all the quantization table / matrix candidate values, and Wse for minimizing mse (i, j) is obtained.
Find the value of (i, j).

The above processing is performed for each (i, j).
Then, finally, the value of W (i, j) giving the minimum value of mse (i, j) is output.

In this embodiment, the difference between the original discrete cosine transform coefficient X _T (i, j) and the inverse quantization value X (i, j) is compared for each coefficient, and the mean square is minimized. Find the quantization table / matrix. For this reason, not only the difference caused by the dynamic range difference but also the difference caused by other factors can be minimized comprehensively.

Therefore, as compared with the case where the fixed quantization table / matrix is used as in the first and second embodiments, the image coded using the reversible discrete cosine transform is converted to the conventional discrete cosine inverse transform. The image quality at the time of decoding using is improved.

[0165]

Embodiment 6 FIG. 6 is a diagram for explaining an embodiment of the sixth invention of the present application, and is a block diagram showing an example of the configuration of a quantization table / matrix generation device for JPEG / MPEG compatibility. is there.

In this embodiment, a sub-optimal quantization table / matrix value is calculated by approximation calculation.
Here, in the case of JPEG, the inverse quantization value of the above equation (19) is approximated by the following equation (43), and in the case of MPEG,
The above equation (40) is approximated by the following equation (44).

[0167]

(Equation 26)

Here, according to the sign of X _q (i, j), ±
As can be seen from the above equations (19) and (40), 0.5 is added in the inverse quantization of JPEG and MPEG. By rounding down or rounding up or down the value after the decimal point.

FIG. 10 schematically shows this. In FIG. 10, a step-like graph f (y) illustrates the above equations (19) and (40), and shows the relationship between values before and after the conversion into integers. This is approximated by a function (the following equation (45)) represented by a straight line in FIG.

[0170]

[Equation 27]

The inverse quantization value X (i, i,
j) is equivalent to the above equations (43) and (44).

By this approximation, e of the above equation (42) is obtained.
(I, j) is the quantization table / matrix value W (i, j)
j) is a linear function. Therefore, the root mean square value mse (i, j) of e (i, j) is a quadratic function of W (i, j), and the quantization table / matrix value W (i, j) for minimizing the quadratic function. j) is obtained analytically.

Quantization table / matrix value W (i,
Considering that j) is an integer, when this value is obtained, J
The following equation (46) is used for PEG, and the following equation (47) is used for MPEG.

[0174]

[Equation 28]

Here, rnd (x) is a function that returns a value obtained by rounding off the decimal part of x, and Σ is the same (i, j) within the cycle (for example, picture) for calculating the quantization table / matrix. On the other hand, it means finding the sum. If the values of the above equations (46) and (47) are out of the range of possible values of the quantization table / matrix value, the nearest integer value within the possible range is set to the quantization table / matrix. Value.

Referring to FIG. 6, according to this embodiment,
2 shows a configuration example of a quantization table / matrix calculation circuit for JPEG / MPEG compatibility.

[0177] With reference to FIG. 6, X '(i, j ) calculating circuit 40, a reversible DCT coefficients X _q (i, j) obtained in the process of dequantizing X' (i, j) the calculate.
This X '(i, j) corresponds to a value obtained by dividing a value before being converted into an integer in the inverse quantization by a quantization table / matrix value. Specifically, in the case of JPEG, X '(i, j) is given by the following equation (48) from the above equation (18).

[0178]

(Equation 29)

In the case of MPEG, from the above equation (39),
The following equation (49) is obtained.

[0180]

[Equation 30]

The X ' _T (i, j) calculation circuit 41 calculates X _q (i, j)
When j) is positive, X _T (i, a value obtained by adding 0.5 to j), when X _q (i, j) is negative, X _T (i, j) from 0.
When X _q (i, j) is 0, the value obtained by subtracting 5 is X _T (i, j).
j) is a circuit in which X ′ _T (i, j) is used.

Quantization table / matrix calculation circuit 42
Is the output X '(i, j) of the X' (i, j) calculation circuit 40.
And the output X ′ _T (i, j) of the X ′ _T (i, j) calculation circuit 41
Then, the value of the following equation (50) is calculated and output as a quantization table / matrix value W (i, j).

[0183]

(Equation 31)

Here, rnd (x) is a function that returns a value obtained by rounding off the decimal part of x. The value obtained by equation (50) is the same as the quantization table / matrix value shown in equation (46) or (47). The inverse quantizer 30 in FIG. 6 is the same as that shown in FIG.

Next, the operation of the circuit of the embodiment shown in FIG. 6 will be described.

The reversible discrete cosine transform coefficient X _q (i, j), which is one of the inputs to this circuit, is converted to X ′ (i, j) in the X ′ (i, j) calculation circuit 40. On the other hand, the discrete cosine transform coefficient X _T (i, j) is calculated by the X ′ _T (i, j) calculation circuit 41 based on the sign of X _q (i, j).
X ′ _T (i, j).

Then, in the quantization table / matrix calculation circuit, the X ′ (i, j) and X ′ _T (i, j)
From j), the quantization table / matrix value W (i, j) shown in the above equation (50) is calculated. This calculation is performed for each (i,
j). If the value obtained by this equation is out of the range of possible values of the quantization table / matrix value, the closest integer in the range is output as the quantization table / matrix value as described above.

Then, in the inverse quantizer 30, the reversible discrete cosine transform coefficient X _q (i, j) is inversely quantized using the value of the quantization table / matrix value W (i, j).
An inverse quantization value is output.

The feature of this embodiment is that a result close to that of the fifth embodiment can be obtained by direct calculation. In the fifth embodiment, e (i, j) is calculated for each quantization table / matrix value to calculate an optimal quantization table / matrix value. Therefore, the processing amount becomes considerably large. In contrast, this sixth
In the embodiment, the amount of calculation can be reduced by directly calculating the quantization table / matrix value by partially approximating the inverse quantization expression.

[0190]

Embodiment 7 FIG. 7 is a diagram for explaining an embodiment of the seventh invention of the present application, and is a block diagram showing an example of the configuration of a quantization table / matrix generation device for JPEG / MPEG compatibility. is there. This embodiment basically has a configuration in which the circuits of the embodiments shown in FIGS. 5 and 6 are combined.

Referring to FIG. 7, the first JPEG / MPE
The G-compatible quantization table / matrix generation device 50
A quantization table / matrix is calculated from the reversible discrete cosine transform coefficient and the discrete cosine transform coefficient, and is output.

The second JPEG / MPEG compatible quantization table / matrix generation device 51 basically has the structure shown in FIG.
The quantization table / matrix is calculated from the reversible discrete cosine transform coefficient and the discrete cosine transform coefficient, and is output. However, the second JPEG / MPE
The G-compatible quantization table / matrix generation device 51 uses the first JPEG / MPEG-compatible quantization table / matrix.
Based on the quantization table / matrix values obtained by the matrix generation device 50, the quantization table / matrix is calculated after limiting the candidates. At the same time, an inverse quantization value of the reversible discrete cosine transform coefficient is obtained.

The feature of this embodiment is that the quantization matrix is obtained in the same manner as in the fifth embodiment, after limiting the candidates based on the quantization matrix value obtained in the sixth embodiment. On the point. By doing this,
In this embodiment, it is possible to obtain almost the same result as in the fifth embodiment with a smaller amount of processing than in the fifth embodiment.

[0194]

Eighth Embodiment FIGS. 8 and 9 are block diagrams showing a configuration of an eighth embodiment of the present invention. FIG. 8 is a diagram showing an 8-ary reversible discrete cosine transform circuit for performing a horizontal operation, and FIG.
8 shows an octal reversible discrete cosine transform circuit performed in the vertical direction. FIGS. 8 and 9 are separated for convenience of drawing.

Referring to FIG. 8, converters 200 to 213 are circuits for performing a reversible quantization after linearly converting the input by the matrix of the above equation (23). Among these, in the reversible quantization from the converter 200 to the converter 209, the quantization cycle is (N ₁ , N ₂ ) and the inverse quantization cycle is (M ₁ , M ₂ ).

[0196] On the other hand, the reversible quantization of transducer 213 from the transducer 210, and the quantization period and _{(N '1, N' 2} ), the inverse quantization period and _{(M '1, M' 2} ). In order to distinguish between them, in FIG. 8, the lossless quantization from the converter 200 to the converter 209 is represented by Q, and the conversion from the converter 210 to the converter 2
The reversible quantization up to 13 is represented by Q '.

Next, the configuration shown in FIG. 8A will be described. The input signal is represented by x (i, j) (i = 0,..., 7, j =
0, ..., and 7), which was the result obtained by 8-way reversible discrete cosine transform in the horizontal direction X _h (i, j) and, X _h (i,
Let _Xq (i, j) be the result obtained by performing an 8-ary reversible discrete cosine transform on j) in the vertical direction.

The converter 200 receives the input signal (x (i, 0),
x (i, 7)) and outputs the result as (u ₀ , u ₄ ). The converter 201 receives the input signal (x (i, 1),
x (i, 6)) and outputs the result as (u ₂ , u ₆ ). The converter 202 receives the input signal (x (i, 3),
x (i, 4)) and outputs the result as (u ₁ , u ₅ ). The converter 203 receives the input signal (x (i, 2),
x (i, 5)) and outputs the result as (u ₃ , u ₇ ).

The converter 210 converts (u ₀ , u ₁ ) and outputs the result as (v ₀ , v ₁ ). Converter 2
11 transforms (u ₂ , u ₃ ) and converts the result to (v ₂ , u ₃ ).
v ₃ ).

The converter 212 converts (v ₀ , v ₂ ) and outputs the result as (X _h (i, 0), X _h (i, 4)). The converter 214 converts (v ₁ , v ₃ ) into the above equation (2)
After conversion by the matrix of 4), lossless quantization is performed, and the result is output as (X _h (i, 2), X _h (i, 6)).

The converter 215 calculates (u ₄ , u ₅ , u ₆ , u ₇ )
Is linearly transformed by a 4 × 4 matrix shown in the above equation (12),
Reversible quantization is performed, and the result is expressed as (X _h (i, 1), X _h (i,
7), output as X _h (i, 3), X _h (i, 5)).

The above is the circuit for performing the 8-ary reversible discrete cosine transform in the horizontal direction. With this circuit, each i
= 0, ..., by performing conversion on _{7, X h (i, j} ) is the result of 8-way reversible discrete cosine transform in the horizontal direction is obtained.

Next, the configuration shown in FIG. 9 will be described.

The converter 204 receives the input signal (X _h (0,
j), X _h (7, j)) and transform the result to (u ′ ₀ ,
u ′ ₄ ). The converter 205 converts the input signal (X _h (1, j), X _h (6, j)) and outputs the result as (u ′ ₂ , u ′ ₆ ). The converter 206
Convert the input signal (X _h (3, j), X _h (4, j))
The result is output as (u ′ ₁ , u ′ ₅ ). Converter 2
07 converts the input signal (X _h (2, j), X _h (5, j)) and outputs the result as (u ′ ₃ , u ′ ₇ ).

The converter 208 converts (u ′ ₀ , u ′ ₁ ) and outputs the result as (v ′ ₀ , v ′ ₁ ). The converter 209 converts (u ' ₂ , u' ₃ ) and outputs the result as (v ' ₂ , v' ₃ ).

The converter 213 converts (v ′ ₀ , v ′ ₂ ) and outputs the result as (X _q (0, j), X _q (4, j)). Transformer 216 transforms (v ′ ₁ , v ′ ₃ ) with the matrix of the above equation (24), performs reversible quantization, and converts the result to (X _q (2, j), X _q (6, j) )).

The converter 217 outputs (u ′ ₄ , u ′ ₅ , u ′ ₆ ,
u ′ ₇ ) is linearly transformed by a 4 × 4 matrix shown in the above equation (12), then reversibly quantized, and the result is expressed as (X _q (1, j), X _q
(7, j), X _q (3, j), X _q (5, j)). The above is the circuit for performing the 8-ary reversible discrete cosine transform in the vertical direction. The conversion is performed on each of j = 0,..., 7 by this circuit to obtain a conversion coefficient X _q (i, j) of the 8 × 8 reversible discrete cosine transform.

Next, the horizontal and vertical directions shown in FIG.
Dynamic range β (0,0) of the DC component X _q (0,0) of the transform coefficient obtained by the original reversible discrete cosine transform circuit
0) will be described. Β (0,0) when the dynamic range of the DC component of the original discrete cosine transform is 1
Is as shown in the following equation (51).

[0209]

(Equation 32)

If β (0,0) is 1, there is no difference in the dynamic range of the DC component between the reversible discrete cosine transform and the original discrete cosine transform. β (0,0) =
In order to be ₁ , the values of N ₁ , N ′ ₁ , M ₁ , and M ′ ₁ need only satisfy the following expression (52).

[0211]

[Equation 33]

The conditions are, for example, N ₁ , N ₂ , M ₁ , M ₂
When the following equation (53) is satisfied during the above, it is satisfied by setting N ′ ₁ and M ′ ₁ to the following equation (54).

[0213]

(Equation 34)

In the above equation (53), M ₁ M _{2 on the} left side is
It corresponds to the number of quantization points in the basic region, and N ₁ N ₂ /
2 corresponds to the number of conversion points in the basic area. That is, when the transformation point and the quantization point correspond one-to-one,
This condition is satisfied.

In order to realize the above equation (54),
Simply, the input to the reversible quantization need only be reversed.
That is, when the converters 200 to 209 shown in FIG. 8 are configured as shown in FIG.
13 to 213 have the configuration shown in FIG. In this case, there is an advantage that a new quantization correspondence table is unnecessary.

As described above, since the values of the quantization period and the inverse quantum period can be selected so that β (0,0) = 1, the 8 × 8 reversible discrete cosine transform and the original 8 × 8 discrete cosine transform are performed. The dynamic range difference between the DC component and the cosine transform can be completely compensated.

As can be seen from the above, what is required to completely compensate for the dynamic range difference of the DC component is that the quantization cycle and the inverse quantization cycle are:
That is, the condition of the above equation (52) is satisfied.

Therefore, when calculating the DC component X _q (0,0) of the 8 × 8 reversible discrete cosine transform, the above equation (2)
Which of three lossless quantizations based on the matrix of 3)
In the reversible quantization of the stage, the quantization period is set to N ′ ₁ , N ′ ₂ ,
It does not depend on whether the inverse quantization period is M ′ ₁ or M ′ ₂ . Therefore, there are circuits other than those shown in FIG. 8 for completely compensating for the dynamic range difference of the DC component, and the number of circuits is ₆ C ₃ = 20 in total.

As described above, by completely compensating for the dynamic range difference of the DC component, the DC component of the discrete cosine transform is considerably close to that of the original discrete cosine transform. Therefore, the image quality when an image encoded using the reversible discrete cosine transform is decoded by the conventional inverse discrete cosine transform is greatly improved.

[0220]

DESCRIPTION OF THE PREFERRED EMBODIMENTS In order to describe the above-mentioned embodiment of the present invention in more detail, an embodiment of the present invention will be described below.

First, an 8 × 8 reversible discrete cosine transform will be described as an embodiment of the present invention. The 8 × 8 reversible discrete cosine transform can be realized by performing an 8-ary reversible discrete cosine transform on a two-dimensional signal divided into 8 × 8 blocks in the horizontal and vertical directions. As the octal reversible discrete cosine transform, each constant value is set as follows using the method described in the patent application filed on August 23, 1996.

First, the values of a ₂ , a ₆ , a ₁ , a ₃ , a ₅ , and a ₇ are 12, ₅ , ₅ , ₄ , ₃ , and ₁ , respectively. Then, the matrix of the following equation (55)

[0223]

(Equation 35)

The quantization cycle of the reversible quantization of (N ₁ ⁽⁰⁾ ,
N ₂ ⁽⁰⁾ ) = (10,14), and set the inverse quantization period to (M ₁ ⁽⁰⁾ , M
₂ ⁽⁰⁾ ) = (7,10). Also, the matrix of the following equation (56)

[0225]

[Equation 36]

In the reversible quantization of ⁽¹⁾ , the quantization period is (N ₁ ⁽¹⁾ , N ₂ ⁽¹⁾ ) = (139, 139), and the inverse quantization period is (M ₁ ⁽¹⁾ , M ₂ ⁽¹⁾ ) = (13,13).

[0227] In the 4 × 4 lossless quantization, when not otherwise stated Thereafter, the transform coefficients X _1, linear quantization width k _1, k ₇ together 7 X _7, when linearly quantizing the global signal The quantization widths L ₃ and L ₅ are both set to 2593/379.

Then, the local quantization is performed on the representative elements (s ₃ , s ₅ ) in the area shown in FIG.
Is defined by assigning the local quantization value shown in.

Next, examples of the first and second aspects of the present invention will be described. In the first and second embodiments of the present invention, a fixed quantization table / matrix value is used, and this value is obtained as follows.

First, from the above equations (25) to (32),
According to the above equation (35), the dynamic range β (i, j) of the transform coefficient of the 8 × 8 reversible discrete cosine transform is calculated. Table 1 shows the results.

Next, a quantization table / matrix is obtained by the above equation (38). Here, the quantization scale is set to s = 1. The resulting table / matrix is given by the following equation (57).

[0232]

(37)

[0233]

[Table 1]

Next, an example in which the values of the quantization widths k ₁ and k ₇ of X ₁ and X ₇ in 4 × 4 reversible quantization are changed. k ₁ , k ₇
Can only be changed by using a quantization table / matrix. In the conventional coding using the reversible discrete cosine transform, no quantization table / matrix is used, so that the dynamic range of the transform coefficient needs to be close to that of the original discrete cosine transform coefficient. The reason is that if the dynamic range deviates, the compatibility is greatly affected. Therefore, there is little freedom in selecting each parameter value, and X
₁ , the values of the quantization widths k ₁ and k ₇ of X ₇ had to be 7.

On the other hand, in this embodiment, since the quantization table / matrix is used, the degree of freedom in determining this value is increased.

Here, as one example, k ₁ = k ₇ = 9
It shows about the case of. At this time, L ₃ = 2593/54
If the local quantization values as shown in FIG. 12 are assigned to the representative elements (s ₃ , s ₅ ) in the area of FIG. 11A assuming that 1, L ₅ = 2593/519, Redundancy can be made smaller than when k ₁ = k ₇ = 7. As a result, the coding efficiency is improved.

When the quantization table / matrix is obtained in the same manner as in the case where k ₁ = k ₇ = 7, the following expression (58) is obtained. By using this quantization table / matrix,
Even when k ₁ = k ₇ = 9, sufficient compatibility can be maintained.

[0238]

(38)

Next, an embodiment of the fifth invention of the present application will be described. In this embodiment, e
A quantization table / matrix that minimizes the mean square of (i, j) is obtained. This average is taken for each cycle of calculating the quantization table / matrix. Here, it is determined on a frame basis.

For example, when k ₁ = k ₇ = 7, when the quantization table / matrix is obtained for the luminance value of the first frame of the image “flower garden”, the following equation (59) is obtained. When this equation is compared with the quantization table / matrix of the above equation (57), more than half the values match, but some values are slightly different. This is because the difference between the transform coefficient of the reversible discrete cosine transform and the original discrete cosine transform coefficient, including the difference generated due to the deviation of the dynamic range, is totally minimized.

As a result, the compatibility with the conventional discrete cosine transform is improved as compared with the case where the quantization table / matrix of the above equation (57) is used.

[0242]

[Equation 39]

Next, an embodiment of the sixth invention of the present application will be described. In this embodiment, the above equation (46) or the above equation (4)
Calculate the quantization table / matrix according to 7).
Here, the quantization table / matrix is calculated for each frame. In the same manner as in the fifth embodiment, k
_{When 1} = k ₇ = 7 and the quantization table / matrix is obtained for the luminance value of the first frame of the image “flower garden”, the following expression (60) is obtained.

[0244]

(Equation 40)

As can be seen from the above equation (60), although approximation is made, more than half of the values match the values of the above equation (59), and other values are different in the range of ± 2. is there.

Next, an embodiment of the seventh invention of the present application will be described. In this embodiment, first, a quantization table / matrix value is obtained by the method of the sixth invention, and values before and after the quantization table / matrix are set as candidates. Then, as in the fifth aspect, a quantization table / matrix value for minimizing the mean square of e (i, j) in the above equation (42) is obtained from the candidates. As can be seen from the comparison between the above equations (59) and (60), the results obtained in the fifth embodiment and the sixth embodiment are not so different from each other. The / matrix candidate values are the value obtained in the sixth embodiment of the present invention and the value ± 1.

Also, here, the quantization matrix is calculated for each frame. Similarly to the fifth and sixth embodiments of the present invention, the quantization matrix obtained for the luminance value of the first frame of the image “flower garden” with k ₁ = k ₇ = 7 is given by the following equation ( 61).

[0248]

[Equation 41]

From this, it can be seen that a quantization matrix almost the same as the above equation (59) can be obtained. It should be noted that the amount of calculation is significantly reduced as compared with the fifth aspect of the present invention due to the reduced search range.

Next, an embodiment of the eighth invention of the present application will be described. In the eighth invention of the present application, in the 8 × 8 reversible discrete cosine transform, among the six reversible quantizations that are passed in the process of calculating the DC component, the quantization period is set to (N
₁ , N ₂ ), the dequantization period is (M ₁ , M ₂ ), the quantization period is (N ′ ₁ , N ′ ₂ ) and the dequantization period is (M ′ ₁ ) for the remaining three stages. , M ' ₂ ).

When these values satisfy the above equation (52), the dynamic range difference of the DC component is completely compensated.

As described above, this condition is such that N ′ ₁ , N ′ ₂ , N ₁ , N ₂ , M ₁ , and M ₂ satisfy the above equation (53).
By making M ′ ₁ and M ′ ₂ as shown in the above equation (54),
It is filled. Here, the following equation (62) is used as a value satisfying the above equation (53).

[0253]

(Equation 42)

Other values satisfying the above equation (53) include those shown in the following equation (63).

[0255]

[Equation 43]

[0256] Alternatively, the above equation (62), a multiple of the value of (63), N ₁ and N _2, M ₁ and is also considered, such as those interchanged M _2, a small size of the quantization correspondence table The value of the equation (62) is good in terms of ending. In addition, where the quantization period and the inverse quantization period are reversed in the six-stage reversible quantization via the calculation of the DC component can be considered as described above, there are 20 possible cases. Followed.

The result of confirming the operation and effect of the eighth embodiment of the present invention will be described below.

The brightness value of the first frame of the image “Flower Garden” is set to 8 as an embodiment of the eighth invention of the present application.
DC component of × 8 reversible discrete cosine transform and original 8 × 8
FIG. 14 shows the distribution of the difference between the DC components of the discrete cosine transform. FIG. 14 also shows a result regarding a conventional 8 × 8 reversible discrete cosine transform that does not simultaneously compensate for a difference in dynamic range.

FIG. 14 shows that the difference between the DC components is significantly reduced by the embodiment of the eighth invention. The mean square value of this difference is 664.94 in the conventional case.
However, in the case of the eighth invention, it was reduced to 0.96.

Next, the results of experiments on compatibility performed using the above-described embodiments of the present invention will be described. In the coding, the lossless discrete cosine transform was used, and in the decoding, the error of the decoded image when the conventional discrete cosine transform was used was examined. The image is “Flower Garden” (720 × 480, 4: 2: 2
Format) 150 frames were used. The encoding basically follows the MPEG-2 algorithm,
-2 to perform the discrete cosine transform with the algorithm
The encoding was performed in place of the above-described reversible discrete cosine transform.

The transform coefficients obtained by the reversible discrete cosine transform are not further quantized, and the quantization width is set to 1. At this time, in terms of the generated code amount, the MPE
It does not meet the specifications of G-2.

In the decoding, a normal MPEG-2 algorithm using a conventional inverse discrete cosine transform is used. Note that the interval between core images (I or P pictures) is M = 3,
The number of frames in the image group (GOP) was N = 15.

Table 2 shows the average square error of the decoded image obtained with respect to the original image over 150 frames. Here, the results obtained by examining the conventional system and the systems of the second, fifth, sixth, and seventh inventions of the present invention in a case where they are used in combination with the eighth invention of the present application and cases in which they are not used are shown.

First, a comparison between the case in which the method is combined with the method of the eighth invention and the case in which the method is not combined shows that the use of the method of the eighth invention significantly improves the decoded image quality.

Next, comparing the conventional system with the second, fifth, sixth, and seventh inventions of the present application, the systems of the second, fifth, sixth, and seventh inventions are all different from the conventional system. It can be seen that the decoded image quality is significantly improved.

In comparison with the second, fifth, sixth, and seventh aspects of the present invention, the scheme of sequentially calculating the quantization matrix is different from that of the second aspect of the present invention using a fixed quantization matrix. In the fifth, sixth, and seventh aspects of the present invention, it can be seen that an improvement is seen.

Further, when the methods of the fifth, sixth and seventh aspects are compared, the fifth aspect of the invention which requires the most computational amount but finds the optimal quantization matrix has the highest decoding image quality. Good. The sixth method of calculating the quantization matrix by approximation calculation has also obtained quite good results, and the seventh method of reducing the amount of calculation by combining the fifth and sixth methods has the same features as the fifth invention. It can be seen that the same good results can be obtained.

The above can be said for the case where the second, fifth, sixth, and seventh inventions of the present application are coded in combination with the eighth invention.

From the above, it can be understood that the use of the quantization matrix and the compensation of the dynamic range difference of the DC component greatly improve the decoded image quality as compared with the conventional method.

[0270]

[Table 2]

[0271]

As described above, according to the present invention, it is possible to correct the dynamic range difference between the reversible discrete cosine transform and the original discrete cosine transform by using the quantization matrix.

For a DC component that cannot be corrected by the quantization matrix, the 6 × 2 × 2
In the three stages of the reversible quantization, the dynamic range is corrected by performing reversible quantization different from the remaining three stages. As a result, compatibility between the reversible discrete cosine transform and the conventional discrete cosine transform is improved. That is, an image encoded using the reversible discrete cosine transform is
Even when decoding using the conventional inverse discrete cosine transform,
Higher image quality than the conventional method can be secured.

[Brief description of the drawings]

FIG. 1 is a block diagram showing an embodiment of an encoder according to the first invention of the present application.

FIG. 2 is a block diagram showing an embodiment of an encoder according to the second invention of the present application.

FIG. 3 is a block diagram showing an embodiment of an encoder according to the third invention of the present application.

FIG. 4 is a block diagram showing an embodiment of an encoder according to a fourth invention of the present application.

FIG. 5 shows JP in the encoder according to the fifth invention of the present application.
FIG. 2 is a diagram illustrating a circuit configuration of an embodiment of an EG / MPEG compatible quantization table / matrix generation device.

FIG. 6 shows JP in the encoder according to the sixth invention of the present application.
2 is a diagram illustrating a circuit configuration of an embodiment of a quantization table / matrix generation device for EG / MPEG compatibility.

FIG. 7 shows a JP in the encoder according to the seventh invention of the present application.
FIG. 2 is a diagram illustrating a circuit configuration of an embodiment of an EG / MPEG compatible quantization table / matrix generation device.

FIG. 8 is a diagram showing a circuit configuration of an embodiment of a portion for performing an 8-ary reversible discrete cosine transform in a horizontal direction in an 8 × 8 reversible discrete cosine transform circuit according to an eighth invention of the present application.

FIG. 9 is a diagram showing a circuit configuration of an embodiment of a portion for performing a vertical 8-way reversible discrete cosine transform in an 8 × 8 reversible discrete cosine transform circuit according to an eighth invention of the present application.

FIG. 10 shows the JPEG of the encoder according to the sixth invention of the present application.
FIG. 6 is a diagram illustrating functions used to approximate the quantization characteristics of MPEG and JPEG in the / MPEG compatible quantization table / matrix generation device.

FIG. 11A is a diagram illustrating an example of a region where a representative element (s ₃ , s ₅ ) exists in reversible quantization based on 4 × 4 matrix transformation of an 8-ary reversible discrete cosine transform circuit. (B)
In the reversible quantization based on the 4 × 4 matrix transform of the 8-ary reversible discrete cosine transform circuit, the existence of this quantized value when the existence range of the representative element (s ₃ , s ₅ ) is as shown in FIG. It is a figure showing an example of a range.

FIG. 12 illustrates a representative element (s ₃ , s ₅ ) in the reversible quantization based on the 4 × 4 matrix transform of the 8-ary reversible discrete cosine transform circuit.
FIG. 12 is a diagram showing an example of the existence range of the quantization value when the existence range of is shown in FIG.

FIG. 13A is a diagram showing a circuit configuration of an embodiment of a linear transformation and a lossless quantization based on a matrix in the eighth invention of the present application. (B) is a figure which shows the circuit structure of one Embodiment of the linear transformation based on a matrix and lossless quantization in 8th invention.

FIG. 14 is a graph showing an example of a distribution of a DC component difference between a reversible discrete cosine transform coefficient and an original discrete cosine transform coefficient in the eighth invention of the present application and the conventional method.

FIG. 15 is a diagram illustrating a configuration example of an 8-ary reversible discrete cosine transform.

FIG. 16 is a diagram illustrating a configuration example of an original 8-ary discrete cosine transform.

[Explanation of symbols]

REFERENCE SIGNS LIST 1 reversible discrete cosine transform circuit 2 variable length encoder 3 adder 4 predictor 5 motion estimating circuit 11 discrete cosine transform circuit 12 quantization table / matrix generation device for JPEG / MPEG compatibility 21 discrete cosine inverse transform circuit 22 predictor 23 Adder 24 Adder 25 JPEG / MPEG compatible quantization table / matrix generation device 30 Inverse quantizer 31 MSE calculation circuit for each coefficient 32 Quantization table / matrix correction device for each coefficient 40 X ′ (i, j) calculation Circuit 41 X ′ _T (i, j) calculation circuit 42 Quantization table / matrix calculation circuit 50 First quantization table / matrix generation device for JPEG / MPEG compatibility 51 Second quantization table / matrix for JPEG / MPEG compatibility Generator 160-168 Converter 180-188 Converter 200-2 17 Converter

Claims (8)

(57) [Claims] 1. A method for encoding an image signal, comprising: means for performing a reversible discrete cosine transform on the image signal to obtain a transform coefficient; and means for storing a JPEG compatible quantization table. Means for performing variable-length coding on the transform coefficients and the quantization table for JPEG compatibility and outputting a coded signal. 2. A method for encoding a moving image signal, comprising: means for estimating a motion vector between a previously encoded image and an image to be encoded to obtain a motion vector; Means for performing motion compensation using a means for creating a predicted image from the previously coded image; means for subtracting the predicted image from the image to be encoded to create a predicted error image; and Means for performing a reversible discrete cosine transform to calculate a transform coefficient; means for storing an MPEG-compatible quantization matrix; and changing the transform coefficient, the MPEG-compatible quantization matrix, and the motion vector. Means for performing long coding and outputting the result. 3. A method for encoding an image signal, comprising: means for performing a reversible discrete cosine transform on the image signal to obtain a reversible discrete cosine transform coefficient; and performing a discrete cosine transform on the image signal. Means for calculating a transform coefficient; means for calculating and outputting a JPEG compatible quantization table that reduces the difference between the inverse quantized value of the reversible discrete cosine transform coefficient and the discrete cosine transform coefficient; Means for performing variable length coding on the cosine transform coefficient and the JPEG compatible quantization table and outputting a coded signal. 4. A method for encoding a moving image signal, comprising: performing motion estimation between an image encoded in the past and an image to be encoded to determine a motion vector; Means for creating a first predicted image from the previously coded image by performing motion compensation using the first predicted image; subtracting the first predicted image from the image to be coded; Means for creating, means for performing a reversible discrete cosine transform on the first prediction error image to calculate a reversible discrete cosine transform coefficient, and a second predictive image subtracted from the encoding target image to obtain a second predictive image. Means for creating a prediction error image; means for performing a discrete cosine transform on the second prediction error signal to calculate a discrete cosine transform coefficient; and an inverse quantization value of the reversible discrete cosine transform coefficient and the discrete Cosine strange Calculates and outputs a quantization matrix for MPEG compatibility such that the difference from the conversion coefficient becomes small, and outputs an inverse quantization value of the reversible discrete cosine transform coefficient when the quantization matrix for MPEG compatibility is used. Means, performing an inverse discrete cosine transform on the inverse quantized value,
Means for calculating a predicted error reproduced image; means for adding the second predicted image to the predicted error reproduced image to create a reproduced image; and performing motion compensation using the motion vector to perform 2. A moving image signal encoding method, comprising: means for creating a second predicted image. 5. The JPEG / MP according to claim 3 or 4, wherein
The means for calculating the quantization table / matrix for EG compatibility converts the reversible discrete cosine transform coefficients into JPEG / MPEG
Means for inversely quantizing based on the values of the compatible quantization table / matrix to obtain an inversely quantized value; and calculating a mean square of a difference between the discrete cosine transform coefficient and the inversely quantized value for each coefficient. 5. The image according to claim 3, further comprising: a unit that corrects and outputs a value of the quantization table / matrix for JPEG / MPEG compatibility based on the root mean square. 6. Signal encoding method. 6. The JPE according to claim 3 or claim 4.
The means for calculating the G / MPEG compatible quantization table / matrix divides the value before being converted into an integer by inverse quantization from the lossless discrete cosine transform coefficient X _q (i, j) by the quantization table / matrix. A value corresponding to the value X '(i, i,
means for the j), the discrete cosine transform coefficients X _T (i, a j), the reversible discrete cosine transform coefficients X _q (i, depending on the value of j), a value obtained by subtracting the number of minute X _'T (I, j), and a calculation based on the second moment about X ′ (i, j) and X ′ _T (i, j),
Means for calculating and outputting a matrix, and means for inversely quantizing the lossless discrete cosine transform coefficient using the quantization table / matrix and outputting an inversely quantized value. The image signal encoding method according to claim 3 or 4. 7. The means for calculating the JPEG / MPEG compatible quantization table / matrix, the means for calculating the JPEG / MPEG compatible quantization table / matrix, and the means according to claim 6. 5. The image encoding method according to claim 3, further comprising: means for calculating a quantization table / matrix for JPEG / MPEG compatibility. 8. A horizontal circuit for a two-dimensional signal of 8 × 8 blocks.
An apparatus for calculating an 8 × 8 block reversible discrete cosine transform coefficient by performing an eight-way discrete cosine transform that is reversible in a direction and a vertical direction , wherein the reversible discrete cosine transform in the horizontal direction and the reversible discrete cosine transform in the vertical direction are performed. Discrete
A first subunit that performs one of the cosine transforms and the other
And a second sub-unit for performing the 8 × 8 block two-dimensional signal transmission.
Extract the m-th row (or m-th column) from the
, X (0), x (1), ..., x (7)
Perform an 8-ary reversible discrete cosine transform on the
(X (0), X (1),..., X (7))
8 × 8 block unidirectional reversible discrete cos
And the m-th row (or m-th column) of the
For each row (or each column) from 1 to 8,
Calculates 8x8 block one-way reversible discrete cosine transform coefficient
And the second subunit is capable of one direction of the 8 × 8 block.
The nth column (or nth row) is calculated from the inverse discrete cosine transform coefficient.
Take out the 8-ary transformed vector signal (x '(0), x'
(1), ..., x '(7)), and this
Octant conversion vector signal (X '(0), X'
(1),..., X ′ (7)).
Let the nth column (or nth row) of the scattered cosine transform coefficient be
This is a row for each column (or each row) from n = 1 to 8.
The 8 × 8 block reversible discrete cosine transform coefficient is calculated.
And the first and second subunits determine whether the elementary signals x (0) and x (7) of the 8-ary transformed vector signal
A binary matrix signal (x (0), x (7)) Is performed, and a reversible quantization is performed to obtain the first 2
A first binary reversible transformation that outputs a source vector signal (u0, u4)
Means for converting the elementary signals x (1) and x (6) of the 8-ary converted vector signal.
To the binary vector signal (x (1), x (6))
Performs a matrix transformation defined by a matrix, performs reversible quantization, and
Output a binary vector signal (u2, u6)
An inverse transforming means for determining whether the elementary signals x (3) and x (4) of the 8-ary transformed vector signal
To the binary vector signal (x (3), x (4))
Performs a matrix transformation determined by a matrix, performs reversible quantization, and
Output a binary vector signal (u1, u5)
An inverse transforming means for determining whether the elementary signals x (2) and x (5)
To the binary vector signal (x (2), x (5))
Matrix transformation determined by the matrix is performed, lossless quantization is performed, and the fourth
Output a binary vector signal (u3, u7)
Inverting means, and an element signal u0 of the first binary vector signal and the third
Vector consisting of the element signal u1 of the binary vector signal of
A matrix transformation defined by the first matrix is applied to the signal (u0, u1).
And a reversible quantization is performed to obtain a fifth binary vector signal (v
0, v1), and a second binary reversible conversion means for outputting the elementary signal u2 of the second binary vector signal and the fourth
Vector consisting of the elementary signal u3 of the binary vector signal of
Matrix conversion determined by the first matrix into the signal (u2, u3)
And a lossless quantization is performed to obtain a sixth binary vector signal.
a sixth binary reversible conversion means for outputting (v2, v3); an element signal v0 of the fifth binary vector signal;
Vector consisting of the elementary signal v2 of the binary vector signal of
Matrix conversion determined by the first matrix into the signal (v0, v2)
And the reversible quantization is performed to obtain the 8-ary transformed vector signal.
A seventh binary vector signal (X (0), X
(4)) a seventh binary reversible conversion means for outputting the elementary signal v1 of the fifth binary vector signal and the sixth binary reversible
Binary vector signal consisting of elementary signal v3 of binary vector signal
(V1, v3), the second matrix consisting of integers a2, a6 Performs a matrix transformation defined by
Eighth binary vector consisting of element signals of the transformed vector signal
Eighth binary reversible converter that outputs the X-ray signals (X (2), X (6))
A stage , an element signal u4 of the first binary vector signal and the third
The elementary signal u5 of the binary vector signal and the second binary vector
Of the elementary signal u6 and the fourth binary vector signal
A quaternary vector signal (u4, u5, u6, u7) consisting of the raw signal u7,
Third matrix of integers a1, a3, a5, a7 Performs a matrix transformation defined by
A quaternary vector signal composed of the element signals of the transformed vector signal
(X (1), X (7), X (3), X (5))
And a step in each of the first and second subunits.
DC component of the 8 × 8 block reversible discrete cosine transform coefficient
The reversible conversion involved in the calculation of the signal, the first said two-way reversible converter of the second two-way reversible transform
Means, said third binary reversible transform means, and said fourth binary reversible means
A first-stage reversible conversion unit including an inverse conversion unit , the fifth binary reversible conversion unit, and the sixth binary reversible conversion unit
Means and a third-stage reversible transform comprising the seventh binary reversible transform means.
Means, the first stage, the second stage, the
A third stage, and the first stage before the second subunit,
The quantization correspondence table used in the reversible quantization is changed between any three of the six reversible transform means of the second and third steps and the remaining three steps. A reversible discrete cosine transform coefficient calculating device , characterized in that a dynamic range difference between a DC component of the reversible discrete cosine transform and a DC component of the original discrete cosine transform is eliminated.