JPH02198273A - Picture encoding system - Google Patents

Abstract

PURPOSE:To prevent the degradation of picture quality by changing a method for truncation to limit an output bit length after orthogonal transformation based on the statistical characteristic quantity of the signal power, etc., of a signal sequence to be inputted. CONSTITUTION:For an original picture signal to be inputted, either the signal power by one screen or a dynamic range to approximately express the signal power is obtained, and corresponding control information is generated by a signal level detecting means 1. In a discrete cosine transforming means 11a, after discrete cosine transformation is executed, a transformation coefficient sequence, for which a bit length is limited by rounding a low-order bit and cutting a high-order bit, is outputted. Here, the bit length, for which the truncation is executed, is set based on the control information obtained by the means 1. The transformation coefficient sequence is quantized by a quantizing means 12, a quantized index is given, the quantized index is encoded in an encoding means 13, the control information is also encoded as additional information, and a code word is obtained.

Description

[Detailed description of the invention] [Industrial application field]

The present invention relates to an image encoding method that performs orthogonal transformation on an image signal and then encodes the signal.

[Conventional technology]

Conventionally, discrete cosine transform coding, extrapolation prediction-discrete sine transform coding, etc. have been known as methods for highly efficient image coding by reducing the redundancy of image signals using orthogonal transform. ing. In these encoding methods, an input image signal is divided into blocks of a predetermined size, and an input image signal sequence is provided for each block. In the discrete cosine transform encoding method, as shown in FIG. The quantization means 12 quantizes and gives a quantization index Q, and the encoding means 13 encodes this quantization index Q. Here, the above block is normally a square block containing NXN pixels, and is generally expressed as a matrix (in the figure, the subscripts are ij), but for the sake of simplifying the explanation, it is expressed as a column vector. It is expressed as. To reproduce an image from a coded signal, the decoding means 21 reproduces the quantization index Q, and the dequantization means 24 reproduces the transformation coefficient sequence (text,
) is reproduced and subjected to inverse discrete cosine transformation by the inverse discrete cosine transform means 25a to obtain a reproduced image signal sequence (it
) is obtained. On the other hand, in the extrapolation prediction-discrete sine transform coding method, the ninth
As shown in the figure, the prediction means 17 predicts the next input image signal sequence based on the reproduced image signal sequence (1) that has already been reproduced after encoding to obtain a predicted image signal sequence (1). The input image signal sequence (χ) is the predicted image signal sequence (foot)
This difference is the prediction error signal sequence (y + l
The discrete sine transformation means 11b performs a discrete sine transformation on the resulting transform coefficient sequence (y, l, and the quantization means 12 gives a quantization index Q. This quantization index Q is then applied to the encoding means. 13. The reproduced image signal sequence (kl
) is a prediction error signal sequence (9, ) and add it to the predicted image signal sequence (leg □) predicted by the prediction means 17. The reproduced image signal sequence (k,) is stored in the delay memory 16 in preparation for future prediction, and the reproduced image signal sequence (k,) stored in the delay memory 16 is used for prediction by the prediction means 17. . To reproduce an image from a coded signal, as shown in FIG. 10, a quantization index Q is reproduced by a decoding means 21, and a transform coefficient sequence ( ), performs inverse discrete sine transformation on the transform coefficient sequence (♀1) by the inverse discrete sine transformation means 24 to reproduce the prediction error signal sequence (夛, ), and furthermore, the prediction means 27
A reproduced image signal sequence (k,) is obtained by adding the predicted image signal sequence (k, ) predicted by . Playback image signal sequence (
k,) is stored in the delay memory 26 in preparation for future prediction, and the reproduced image signal sequence (k,) stored in the delay memory 26 is
kl) is used for prediction by the prediction means 27. Here, except for the decoding means 21, the reproduction of the image is the same as the configuration in which the predicted image signal sequence (leg) was obtained during encoding.

[Problem to be solved by the invention]

As described above, in any encoding method, in order to reduce the redundancy of the input image signal sequence, the input image signal sequence is subjected to orthogonal transformation expressed by the following equation. [X,]=[a,,][χ□] Here, Xl is a component of the transformation coefficient sequence, χ is a component of the input image signal sequence or prediction error signal sequence, and a and J are the components of the orthogonal transformation matrix. be. When such orthogonal transformation is performed, the difference between the maximum and minimum values of the components becomes larger in the matrix after transformation than in the matrix before transformation. That is, since the dynamic range increases due to orthogonal transformation, the output bit length becomes larger than the input bit length. Also, in the above explanation, the input image signal sequence was explained as being one-dimensional, but since images have two-dimensional correlation, the input image signal sequence is actually divided into two-dimensional blocks as shown in Figure 11. The orthogonal transformation for the input image signal sequence is a two-dimensional orthogonal transformation. In calculations in two-dimensional orthogonal transformation, one-dimensional orthogonal transformation in the column direction (horizontal direction of the matrix) is performed a number of times equal to the number of columns, and then one-dimensional orthogonal transformation in the row direction (vertical direction of the matrix) is performed the number of times equal to the number of rows. Since the orthogonal transformation is performed twice in this way, the final output bit length becomes considerably larger than the input bit length. Therefore, in order to simplify the hardware by reducing the bit length of data to be processed, it is necessary to limit the bit length of data. To limit the bit length, as shown in FIG. 12, it is possible to round off the lower bits or cut the upper bits. That is, FIG. 12 shows an example in which the dynamic range of the input image signal sequence is 10 bits long, and the dynamic range becomes 15 bits long by performing orthogonal transformation, but is limited to 10 bits long. Figure 12 (
In a), the output is limited to 10 bits by rounding the lower 5 bits, and in Fig. 12(b), the output is limited to 10 bits by rounding the lower 4 bits and cutting the upper 1 bit. It is limited in length. In this way, when the lower bits are rounded or the upper bits are cut, a truncation error occurs. That is, when processing as shown in FIG. 12(a), there is a problem in that the lower bits are rounded, resulting in a large rounding error. On the other hand, when processing as shown in Fig. 12(b), the rounding error for the lower bits becomes smaller than in the case of Fig. 12 (st>), and since only one bit is cut from the above bits, the input When the signal power of the image signal sequence is relatively small, the error is small. However, when the signal power of the input image signal sequence is relatively large, many errors occur due to the cutting of the upper bits, which affects the image quality of the reproduced image. A problem arises in that the input image signal sequence is generally diverse, so the 12th
If the output bit length is limited by either the processing shown in FIG. 12(a) or FIG. 12(b), a problem arises in that it cannot cope with the diversity of input image signal sequences. The present invention aims to solve the above problems, and provides an image encoding method that prevents deterioration of image quality due to truncation errors while limiting the output after orthogonal transformation to a predetermined bit length. This is what I am trying to do.

[Means to solve the problem]

In order to achieve the above object, the present invention switches the method of limiting the output bit length after orthogonal transformation depending on the characteristic quantity such as the signal power of the input signal sequence. Alternatively, the original image may be divided into a plurality of regions, and the method for limiting the output bit length after orthogonal transformation may be switched for each region based on a characteristic quantity such as signal power.

[Effect]

According to the above configuration, the method for limiting the output bit length after orthogonal transformation (hereinafter referred to as the truncation method) is changed based on the characteristic quantity such as the signal power of the input signal sequence, so after orthogonal transformation, When limiting the output bit length, the optimal truncation method can be selected according to the input signal sequence, making it possible to almost eliminate truncation errors and suppress deterioration of image quality due to truncation errors. This is possible. Furthermore, by dividing the original image into multiple regions and switching the truncation method for each region, encoding can be performed in response to local changes in image characteristics, further improving image quality. Deterioration can be suppressed.

Embodiment 1 FIG. 1 shows an embodiment in which the present invention is applied to a discrete cosine transform encoding method. For the input original image signal, the signal level detection means 1 determines the total value of signal power (pixel brightness) for one screen or the difference between the maximum value and minimum value of signal power (dynamic range), Corresponding control information is generated. Next, one screen is divided into a plurality of blocks having a predetermined size, and each block is input as an input image signal sequence to the conventional discrete cosine transform means 11a. In the discrete cosine transform means 11a, after performing the discrete cosine transform,
Outputs a transform coefficient sequence with a limited bit length by rounding the lower bits or cutting the upper bits. Here,
The bit length for truncation is set based on control information obtained from the signal level detection means 1. The outputted transform coefficient sequence is quantized by the quantization means 12 and given a quantization index, and the quantization index is encoded by the encoding means 13. Here, since the bit length of the transform coefficient sequence by the discrete cosine transform means 11a changes depending on the signal level of the input image signal sequence, the scale of the transform coefficient sequence also changes. Therefore, the scale of the quantization step width is also adjusted according to the control information generated by the signal level detection means 1. In the encoding means 3, the control information is also encoded as additional information along with the quantization index, and a code word is obtained. In order to reproduce an image signal from the code word, the decoding means 2
1, the quantization index and control information are reproduced from the code word, and the dequantization means 24 sets the scale of the quantization step width based on the control information, and then performs dequantization and reproduces the transform coefficient sequence. . Next, the inverse discrete cosine transform means 25a restores the truncated bit number according to the control information and performs inverse discrete cosine transform to obtain a reproduced image signal. In the above embodiment, the signal power of the original image signal is measured for each screen, but the original image signal is divided into multiple areas including multiple blocks and the signal power is measured for each area. , the signal level of the original image signal can be adjusted in units of areas, or one area can be configured with one block, the signal power can be measured in units of blocks, and the signal level of the original image signals can be adjusted in units of blocks. Good too. When adjusting the level in units of blocks, since the signal input to the signal level detecting means 1 is divided into blocks, there is no need to divide the signal into blocks after adjusting the signal level. Also,
By adjusting the signal level on a region-by-area or block-by-block basis, it is possible to perform encoding that adapts to local changes in signal power of the image. Moreover, the signal level detection means 1
Since the processing by the signal level adjusting means 2 and the signal level adjusting means 2 can be performed in parallel, pipeline processing is possible, which leads to a reduction in processing time. However, since the control information also needs to be encoded as additional information for each region or each block, it is desirable to compress the additional information by run-length encoding the control information.

Embodiment 2 In this embodiment, as shown in FIG. 2, the present invention is applied to an extrapolation predictive discrete sine transform coding method. In the extrapolation prediction-discrete sine transform encoding method, an input image signal sequence obtained by dividing an original image signal into a plurality of blocks is input. The input to the discrete sine conversion unit 7 is a prediction error signal sequence that is the difference between the input image signal sequence and the predicted image signal sequence predicted by the prediction means 17 based on the reproduced image signal sequence that has been quantized and already reproduced. Therefore, the signal level detection means 1 generates control information in the same way as in the first embodiment, based on characteristic quantities (for example, correlation coefficients, etc.) that reflect the signal power of the prediction error signal sequence. The discrete sine transformation means 11b performs discrete sine transformation on the input image signal sequence by changing the method of truncating the output transform coefficient sequence according to the control information generated by the signal level detection means 1,
Output the conversion coefficient sequence. The quantization means 12 performs quantization by setting a step width scale for the transform coefficient sequence based on the control information, and outputs a quantization index. The encoding means 13 encodes the quantization index into a variable length code, encodes the control information generated by the signal level detection means 1 as additional information, and outputs a code word. Incidentally, the quantization index is dequantized by the dequantization means 14 using a step width whose scale is set based on the above control information to reproduce a transform coefficient sequence, and the transform coefficient sequence is subjected to inverse discrete sine transformation. The means 15 restores the truncated number of bits based on the control information and performs inverse discrete sine transformation to reproduce the prediction error signal sequence. The prediction error signal sequence is transmitted to the prediction means 17
A reproduced image signal sequence is obtained by adding the predicted signal sequence from , and the reproduced image signal sequence is stored in the delay memory 16 and used for subsequent prediction. In order to reproduce an image signal from a code word, as shown in FIG. Inverse quantization is performed using the step width with the scale set to reproduce the transform coefficient sequence. Further, the inverse discrete sine transform means 25
After restoring the number of bits truncated according to the control information and performing inverse discrete sine transformation to reproduce the prediction error signal sequence in step b, the prediction error signal sequence is added to the prediction signal sequence from the prediction means 27 to obtain the reproduced image signal. You get it. Further, the reproduced image signal sequence is stored in the delay memory 26 and used for subsequent prediction. FIG. 4 shows the relationship between the amount of code and the SN ratio for evaluating how to switch the number of truncation bits in the first and second embodiments. Here, the truncation method is performed by switching the processes shown in Figures 12(a) and (b) (the line passing through the O mark in the figure is the 12th
(corresponds to Figure 12(a), and the line passing through the mouth corresponds to Figure 12(b)), and the input image is a general natural image (
The signal power of the prediction error signal is relatively small because the correlation with the original image signal is strong, and the truncation method is ranked high in both the methods shown in Fig. 12 (a) and Fig. 12 (b). No bit cutting occurs. Therefore, the method shown in FIG. 12(b), which has a smaller rounding error, has a higher signal-to-noise ratio. In addition, for images with strong contrast that include printed characters, the signal power of the prediction error signal may increase partially as shown in the table below, so the method shown in Figure 12 (b) is better for cutting out the upper bits. A large number of errors occur, and the S/N ratio is lower than the method shown in Figure 12 (a) due to the influence of errors caused by cutting out the bits mentioned above.

Embodiment 3 In this embodiment, another example in which the present invention is applied to an extrapolation prediction/discrete sine transform encoding method will be described. The input signal sequence is a block of a predetermined size obtained by dividing the original image signal, and the prediction error signal sequence is subjected to discrete sine transformation by the discrete sine transformation means 11b. Control information for switching the truncation method in the N-dispersion in conversion means 11b is obtained based on the prediction error signal reproduced for the already encoded block. That is, as shown in FIG. 7, by referring to the signal power of the prediction error signal reproduced for blocks A to D that have already been encoded around the block X for the block X to be encoded, The method of truncating block X is determined according to certain rules such as majority voting or arithmetic mean. Block A that has already been encoded in the vicinity of block X that is the target of encoding in this way
Since the control information for the target block Not only is it possible to perform switching that adapts to the local characteristics of the code, but since the encoded information is referenced, there is no need to encode new control information, and no additional information is generated in the code word. It is. The signal level detection means 1 measures the signal power of the reproduced prediction error signal of the encoded block, generates control information indicating the magnitude of the signal power, and stores it in the storage means 18. The storage means 18 stores information on the signal power of encoded blocks. The discrete sign conversion means 11b determines a truncation method by referring to information on blocks surrounding the block to be encoded among the information stored in the storage means 18, and generates control information based on this. Then, the discrete sine transform is performed. To reproduce an image signal from a code word, as shown in FIG. 6, the signal power of the prediction error signal in the already reproduced block is measured by the signal level detection means 1b, and the control information based on this signal power is stored in the storage means. The number of bits to be truncated is determined by referring to the information of blocks surrounding the target block stored in the block 29, control information is generated, and discrete sine conversion is performed based on this control information. The other configurations are the same as those in the second embodiment, so their explanation will be omitted.

【Effect of the invention】

As described above, the present invention switches the method of limiting the output bit length after orthogonal transformation according to the characteristic quantity such as the signal power of the input signal string, and Since the truncation method is changed based on such characteristic quantities, the optimal truncation method can be selected according to the input signal sequence when limiting the output bit length after orthogonal transformation, and truncation errors can be reduced. This has the advantage that it can be prevented from occurring, and deterioration in image quality due to truncation errors can be suppressed. Furthermore, by dividing the original image into multiple regions and selecting a truncation method for each region, encoding can be performed that corresponds to changes in local characteristics of the image, further reducing image quality deterioration. It has the effect of suppressing

[Brief explanation of the drawing]

FIG. 1 is a schematic block diagram showing a first embodiment of the present invention, FIG. 2 is a schematic block diagram of a part that performs encoding processing, showing a second embodiment of the present invention, and FIG. 3 is a block diagram showing the same decoding process. 4 is an explanatory diagram of the operation of the present invention; FIG. 5 is a schematic diagram of a portion that performs encoding processing according to the third embodiment of the present invention; FIG.
The figure is a schematic block diagram of the decoding part of the same as above, FIG. 7 is an explanatory diagram showing the concept of the process of obtaining control information in the above, FIG. 8 is a schematic block diagram of a conventional example, and FIG. 9 is a diagram of another A schematic configuration diagram showing a conventional example, FIG. 10 is a schematic configuration diagram of a part that performs decoding processing in the same as above, FIG. 11 is an operation explanatory diagram showing the concept of the calculation process of orthogonal transform in the same as above, and FIG. FIG. 2 is an explanatory diagram showing an example of a truncation method. 1.1b...Signal level detection means, 2...Signal level adjustment means, lla...Discrete cosine transformation means, 11
b...Discrete sine conversion means, 12...Quantization means,
13... Encoding means, 14... Inverse quantization means, 15
... Inverse discrete sine conversion means, 16... Delay memory,
17... Prediction means, 18... Signal level reverse adjustment means, one... Storage means, 21... Decoding means, 24.
... Inverse quantization means, 25a... Inverse discrete cosine transform means, 25b... Inverse discrete sine transform means, 26... Delay memory, 27... Prediction means, 28... Signal level inverse adjustment means , 29...memory means. Agent Patent Attorney Ishi 1) Chief 7 Figure 8 BA and 9 Figure 10 Voluntary writing of procedural amendments) 1. Display of the case 1999 Patent Application No. 17305 2, Name of the invention Image encoding method 3, Person making the amendment Relationship to the case Patent applicant address 1048 Kadoma, Kadoma City, Osaka Prefecture Name (58
B) Matsushita Electric Works Co., Ltd. Representative: Toshio Miyoshi 4, Agent postal code: 530 5, Date of amendment order: 6, Number of claims increased by amendment: None 7, Specification and drawings subject to amendment [1] The claims in the specification of this application are amended as follows. [(1) In an image encoding method that performs orthogonal transformation on a signal sequence obtained from an image signal and then encodes the signal sequence, the value after orthogonal transformation is determined according to the signal power characteristics of the input signal sequence. An image encoding method characterized by switching the method of limiting the output bit length. (2) The original image is divided into a plurality of regions, and the method for limiting the output bit length after orthogonal transformation is switched based on the signal power 1a characteristic quantity for each region. ``Image encoding system,'' [2] ``Characteristic quantities such as signal power'' on page 8, line 4 and page 8, line 8 of the same are corrected to ``statistical characteristic quantities such as signal power,'' respectively. . [3] Delete the entire text of page 9, line 11 to line 13 of the same as above, and insert the following sentence. 1 determines the signal power for one screen (the root mean square value of pixel brightness) or the dynamic range that approximately expresses this (the difference between the maximum and minimum pixel brightness values), and [4] "Encoding means 3" on page 10, line 12 of the above is corrected to "setting the bit length for encoding" and "encoding means 13". [5] In the 1st and 2nd lines of page 11, ``Restore the number of truncated bits'' is corrected to ``Set the bit length for r-truncation''. [6] Same as above, page 11, lines 7 to 8, page 11, line 10
"Adjust the signal level of the original image signal" in lines 1 to 11 is corrected to "change the truncation method." [7. Correct ``Adjust the signal level'' on page 11, line 16 of the same page to ``change the truncation method.'' [8] Delete the words "Mosakamo" on page 11, line 18 of the same as above to "desu" on page 12, line 1. [10] On page 13, lines 14 and 15 of the same page, the phrase "the number of truncated bits is restored" is corrected to "the bit length for truncation is set." [11] "Restore the truncated number of bits and 11" in the 7th to 8th lines of the 14th page of the same page. [12] The 1st to 2nd lines of the 18th page of the same page ``Correct the truncated J to ``perform truncation.'' [13] Same as above, page 18, lines 7 to 8, page 18, line 1
Correct "characteristic quantity such as signal power" in line 0 to "statistical characteristic quantity such as signal power." [14] Delete "Signal level inverse adjustment means, 19" on page 20, line 3 of the same. [15] Delete "28...Signal level reverse adjustment means" on page 20, line 7 of the same. [Of the 16 attached drawings, Figures 1, 2, 3, 5,
Figures 6 and 9 are corrected as shown in the attached sheet. Agent Patent Attorney Ishi 1) Choshichi

Claims (2)

[Claims] (1) In an image encoding method in which a signal sequence obtained from an image signal is encoded after orthogonal transformation, the output bit length after orthogonal transformation is determined according to the characteristic quantity such as the signal power of the input signal sequence. An image encoding method characterized by switching a method of limiting. (2) The original image is divided into a plurality of regions, and the method of limiting the output bit length after orthogonal transformation is switched for each region based on a characteristic quantity such as signal power. Image encoding method.