JPH03253119A - Control information reducing system in adaptive forecasting coding - Google Patents

Abstract

PURPOSE:To improve signal transmission efficiency by inhibiting the transmission of control information for selecting a forecasting function to the decoding side when the coding side can decode a signal without using the control information. CONSTITUTION:The coding side detects a case capable of decoding a signal without using the control information for selecting the forecasting function, and at the time of detecting the case, inhibits the transmission of the control information to the decoding side. When the control information is not received, the decoding side extracts an optimum forecasting function from a forecasting function formed from reproduced picture elements on the decoding side itself in the same procedure as that of the coding side and a transmitted quantized representative value and reproduces each picture element. Thereby, the decoding side can reproduce the picture element by selecting the optimum forecasting function corresponding to each picture element even if the control information is not transmitted. Since the volume of control information to be transferred from the coding side to the decoding side can be reduced, the signal transmission efficiency can be improved.

Description

DETAILED DESCRIPTION OF THE INVENTION (Field of Industrial Application) The present invention relates to the field of predictive coding for compressing image information in digital video conferences, video telephones, and the like.

(Prior art) In conventional adaptive predictive coding methods, the optimal prediction function is selected for each pixel, and although it is possible to transmit images with fine-grained quality for each pixel, control information of the prediction function is not transmitted for each pixel. The disadvantage is that the signal transmission efficiency is poor because of the need for

(Problems to be Solved by the Invention) Regarding the above point, there are cases where decoding can be performed without control information on the receiving side due to the relationship between the quantized representative value sent from the transmitting side and the prediction function created from the image data reproduction value. If it exists,
It is possible to reduce control information and improve signal transmission efficiency. To this end, the present invention aims to improve signal transmission efficiency by detecting all cases where decoding is possible without control information and performing predictive encoding and decoding based on the detected cases.

(Means for Solving the Problems) In the present invention, as a means for detecting all cases in which decoding is possible without control information and performing predictive encoding and decoding based on the detected cases, several prediction methods are used in an adaptive predictive encoding method. The method is characterized in that a prediction error is minimized and control information is reduced by preparing a function and using the prediction function closest to the value of the pixel to be encoded as a prediction value.

That is, in predictive coding, the prediction function that takes the value closest to the value of the pixel to be encoded is selected as the optimal prediction function from among multiple prediction functions prepared in advance as the prediction function of the pixel to be encoded, and the quantization representative In an adaptive predictive coding method that transmits control information for selecting values and prediction functions to the decoding side and reproduces pixels on the decoding side, the value of the prediction function created using the pixel values reproduced on the encoding side. Detects cases where decoding is possible without using control information for prediction function selection based on the relationship with the quantization representative value, and at the time of detection on the encoding side, transmits the control information for prediction function selection to the decoding side. If the decoding side does not receive control information for selecting a prediction function from the encoding side, the prediction function created from the reproduced pixels on the decoding side and the transmitted quantized This control information reduction method in adaptive predictive coding is characterized in that control information for selecting a predictive function can be omitted by extracting an optimal predictive function from representative values and regenerating pixels.

(Operation) On the encoding side, if decoding can be performed without using control information for prediction function selection, the encoding side does not transmit the control information to the decoding side. Therefore, signal transmission efficiency is improved.

On the decoding side, if no control information is received, the optimal prediction function is extracted from the prediction function created from the reproduced pixels on the decoding side and the transmitted quantized representative value using the same procedure as the encoding side, and the pixel is Play.

Therefore, even though no control information is transmitted, the decoding side can select the optimal prediction function for each pixel and reproduce the pixel.

(Principle of the invention) The principle of the present invention will be explained as follows.

In pixel-by-pixel adaptive prediction, it is assumed that n prediction functions are prepared in advance.

These prediction functions and pixel values to be encoded are expressed as follows.

n prediction functions: f l Hf 2 g 00. l
f n Pixel value to be encoded: X Prediction function L (CI, 2., , ,
n ) is selected as the optimal prediction function.

lx-fll=min(lx-fll, 1x-fll,
-+1x-fll) (1) When decoding, f I + f2+ 30. is used as a prediction function. + fn is selected with equal probability, at least 1og as control information
An amount of information equal to zn is required for decoding. However, at the time of decoding, the quantization representative value at the time of encoding (hereinafter, q
m) and the prediction function f+.

f210. −1 A prediction function that has no possibility of being selected during encoding due to the relationship of the value of fn, that is, a non-optimal prediction function f.

There are cases where (j·1, 2., , n; j≠i) can be detected without control information. This means that the amount of control information passed from the encoding side to the decoding side can be reduced.

The principle by which detection can be performed without the above-mentioned control information will be shown below.

In Figure 3, on the encoding side, the value of the pixel to be encoded
is the prediction function fl, lq+ l≦IX−f, l≦q
, I is selected as the quantization representative value (the value closest to zero in a certain quantization range is ql,
Let the farthest value be qo).

Here, if the following equation (2) holds during decoding, f.

cannot be the optimal prediction function.

Yofk(k=1.2.,,,n; k ≠j)l
fJ"ql-fkl < lq+ 1 (q+≠O)
(2) The reasons are explained below.

If fJ is assumed to be the optimal prediction function, the deviation from X is <lq+1ql as shown in the following equation (3).

If=-XI≧lqt l
(3) Furthermore, since the deviation between X and f should be the smallest, the following equation (4) holds true.

v fklx-fxl≧IX-fJ +
(4) Transform equation (2) into the following equation (5)-1, and (5)
-Two equations can be derived.

(fJ+q1-fk)2Q+"<o
(5)-1(f,-f,) (fi-f,,+2q,)
< O(5)-2 Transform formula (3) into the following formula (6)-1,
Equation (6)-2 can be derived.

(fJ-X)”-ql”≧O(6)-1(f=-X-q
l) (L-X+q+) ≧O(6) -2 Equation (4) is transformed into the following equation (7)-1, and equation (7)-2 can be derived.

(x-fm) 2- (X-ft)” ≧0(f,-f
, l) Assuming (2X-f, -f,)≧Oql〉0, from equation (5)-2, the following equation (8) f12q+ < f, + < fk (6) From equation-2, the following equation (9) can be said.

The following can be said.

(7)-1 (7)-2 (8) X≦f, -ql, X≧fr+q+ (
9) In equation (7)-2, equation (lO) can be said for X from equation (8).

2X-f1fJ≦O,', 2X≦'fk+L
(10) Using equation (9), for f4, (11)
This can be said about the expression.

2q + +2fJ≦2X≦fk+f.

,', f, ≦fh−2qi
(11) This contradicts equation (8).

Also, assuming that ql<0 From equation (5)-2, the following equation (12) can be established.

fk<fJ< fy-2Q+
(12) From equation (6), the following equation (13) can be stated.

X≦f, +ql, X≧f, -ql
(13) In equation (7)-2, from the fact of equation (12), (
14) Equation can be derived.

2X≧fk+fJ Using fk+f, ≦2fJ+2q.

fJ≧f, -2q,
(14) Equation (14) contradicts equation (13).

Therefore, when it is assumed that f, which satisfies equation (2), is an optimal prediction function, it cannot be the optimal prediction function because it causes a contradiction, and it can be determined without control information that f is a non-optimal prediction function.

Note that if fI cannot be specified using equation (2), it must be determined using control information.

That is, this is a case where control information is required.

Furthermore, even when f satisfies the conditions of the following equations (15) and (16), f cannot be the optimal prediction function.

fJ+qi>upper limit of quantization dynamic range (15
)f, +ql<lower limit of quantization dynamic range (1
6) That is, the reproduction value created using the non-optimal prediction function (fJ) becomes an impossible value as image data. Therefore, fJ can be determined to be a non-optimal prediction function.

The above principle will be explained as follows using an example using numerical values.

X=100 f, = 90 f2 = 75 Assume that the following quantization table is used.

Quantization range Representative value (q-)q+ O~2 1 0 3~7 5 3 8~14 11 8+X-fll
=10<25= IX-f,ltea6(M-The selected function is fI, qm=11, and q+=8.

At this time, assuming that fi is the optimal prediction function, lfz
+q+-fI 1 = 175+8-901 =7<8, which causes a contradiction, so it can be determined without control information that fl is the optimal prediction function.

(Embodiment) FIGS. 1 and 2 are block diagrams of an image encoding/decoding apparatus to which the present invention is applied.

The functions of each part and the flow of data will be explained in order.

On the transmitting side of FIG. 1, original image data X is input into an image data memory 2 and stored. The prediction functions of each part from prediction function part (1) 3 to prediction function part (n) 5 are calculated by subtractors 12 to 1.
4, the difference with the original image data X is taken to obtain prediction errors 1 to n, which are input to the function comparison section 10.

The function comparison unit 10 finds the prediction function with the smallest difference from the pixel The function is output to the subtractor 18.

The subtracter 18 takes the difference between the original image data X and the optimal prediction function, and inputs the difference to the quantizer 1. Quantization section 1
Then, quantization representative values qm and qI are determined using a quantization table or the like, and qm is set as the output on the transmitting side.

q is input to the control information presence/absence determination unit 6, and it is determined whether the above equation (2) holds true with the values of the prediction function storage unit (1) 3 to the prediction function storage unit (n) 5, that is, the prediction function of each part. is checked, and if it holds true, the switcher 11 is controlled so as not to transmit the prediction function selection information to the receiving side as control information, and if it does not hold, the prediction function selection information is outputted as control information from the transmission side.

Quantization representative value q. When output, the adder 15 locally encodes it with the prediction function selection information and stores it in the image data memory 2 as reproduced image data X'. A prediction function is created in a prediction function creation section 9 based on the accumulated reproduced image data X°, and the prediction function is sequentially stored in prediction function sections (1) 3 to (n) 5, and its contents are updated.

With this, the encoding operation for a certain pixel is completed, and the next encoding process begins.

On the receiving side in Fig. 2, decoding processing is performed based on the quantization representative values q, R and control information sent from the transmitting side to create reproduced image data X°, an outline of which is as follows. That's right.

The quantization decoding unit 101 performs decoding processing on the quantization representative value q1. Further, the control information is processed by the control information presence/absence determination unit 106. Control information presence/absence determination unit 1
The recognition of the presence or absence of control information in step 06 is based on q selected by the quantization decoder 101 based on the quantization representative value qm, and the prediction function section (
From 1003 to prediction function section (n) 105, the presence or absence of control information is determined from the prediction functions of each section according to equation (2). As a result, if it is determined that there is control information, the quantization decoder 101 is notified of the bit value 1ogzn necessary to express no from the group 106, assuming that the number of prediction functions at that time is no. The quantization decoding section 101 extracts log2n' worth of control information from the bit string after the quantization representative value qm and outputs it to the control information presence/absence determining section 106. The same county 106
Now let this value be the prediction function selection information. For a prediction function determined to have no control information, the control information determination unit 106 uses the calculated value as prediction function selection information.

From the above, the selection information (prediction function selection information) that can completely specify the optimal prediction function value f+ is output from the control information presence/absence determination unit 106, and the switcher 107 is controlled thereby, and the prediction function that is the same as that on the transmitting side, that is, the prediction Function part (1) 1
03 to the prediction function section (n) 105 to obtain the optimum prediction function value f.

Note that the position on the input bit string of the quantized representative value to be decoded in the G-th order in the quantization decoding section 101 is recognized by the control information bit value login' obtained from the control information presence/absence determination calculation result.

The optimal prediction function value obtained in this way and the quantized representative value decoded by the quantization decoder 101 are processed by the adder 108 to obtain reproduced image data X°, and the image data memory 1
02 and displayed on a VDU or the like from the image data memory 102. In addition, a prediction function creation unit 109 creates a prediction function based on the reproduced image data do.

(Effect of the invention) As a prediction function of the pixel X to be encoded, horizontally and vertically adjacent pixel values A and B are used as shown in FIG. 1 when adaptive prediction is applied to some test data images using the standard quantization table of
The average amount of control information per pixel was calculated.

The calculation was performed by selecting the optimal predicted pixel and determining the presence or absence of control information according to the following algorithm.

Selection of optimal predicted pixel lX-Al≦lX-B1-predicted pixel X=AIX-AI>
lX-B1→Predicted pixel None, (Presence/Absence, A-B) A-B) A-B) A-B) A-B) A-B) Table 1 3-bit quantization table (quantization) ~26 ~10 ~2 ~- 4 ~-12 ~-28 ~-255 Table 2 4bit Quantization Table (Quantization) 120~150 89~119 58~88 27~57 11~26 3~10 0~2 -1~-4 -5~ -12 -13 to -28 -29 to -59 -60 to -90 -91 to -121 -122 to -152 -153 to -255 35 04 3 2 8 2 9 20 to 44 75 106 137 168 20 9 8 7 1 1 5 13 29 60 91 122 153 3rd table bit Relefted quantization table (quantization) 3 4 0 2 0 2g 66 94 16 32 43 50 54 254 250 243 232 216 194 166 128 90 ~5 ~12 ~2 3 ~ 39 ~61 ~89 ~127 ~165 ~193 ~215 ~231 ~242 ~249 ~253 ~255 ~-255 ~-253 ~-249 ~-242 ~-231 ~-215 ~-193 ~-165 ~-127 7 0 9 4 06 44 78 03 22 36 45 51 54 254 251 245 236 222 203 178 144 106 3 4 0 2 0 28 66 94 16 32 43 50 54 254 250 243 232 216 194 166 128 90 As explained above in detail According to the present invention, the neighboring two
When using pixel values as prediction functions, four natural still images and three test patterns ZONEI (developed by BBC, standard of CBD Revision Society), MB
Table 4 shows the amount of control information per pixel when using the method of the present invention compared to the conventional method I BIT (CCIR standard).

As shown in Table 4, approximately 0.51 to 0 per pixel,
This provides the advantage of being able to perform a fine-grained adaptive predictive coding method on a pixel-by-pixel basis using approximately 63 bits.

That is, the BIT reduction rate in control information is about 37 to 49% compared to the conventional method. Note that in Table 4, when the test pattern is MB, the reduction rate is low, but the probability that such a pattern will actually occur is small, and the overall effect is small.

Below margin 2g -13~-23 29-6~-12 30-2~-5 17 8 3 13 6 2 4th Table 103, 104.105... Prediction function section, 106...
- Control information presence/absence determination unit, 107... Switcher 108 109... Prediction function creation unit.

...adder,

Claims (1)

[Claims] A prediction function that takes a value closest to the value of a pixel to be encoded is selected as an optimal prediction function from among a plurality of prediction functions prepared in advance as a prediction function for a pixel to be encoded, and a quantized representative value is selected. In an adaptive predictive coding method in which control information for selecting a prediction function is transmitted to the decoding side and pixels are reproduced on the decoding side, the value of the prediction function created using the pixel values reproduced on the encoding side and the , detects a case where decoding is possible without using control information for prediction function selection from the relationship with the quantization representative value, and transmits the control information for prediction function selection to the decoding side at the time of the detection on the encoding side. First, if the decoding side does not receive control information for selecting a prediction function from the encoding side, the prediction function created from the reproduced pixels on the decoding side and the transmitted quantization representative are used in the same way as the encoding side. A method for reducing control information in adaptive predictive coding, characterized by omitting transmission of control information for selecting a predictive function by extracting an optimal predictive function from values and regenerating pixels.