JPH0358672A - Method and apparatus for coding picture signal - Google Patents

Abstract

PURPOSE:To obtain a decoded picture with high quality by applying prediction coding completed in a block and varying the quantization characteristic for each block. CONSTITUTION:A prediction error calculation processing section 1 calculates a prediction error signal outputted from an in-block prediction processing unit 5 to an original picture signal. On the other hand, a prediction error dynamic range calculation processing section 6 obtains the dynamic range of the prediction error for each block from an original picture signal. Then a quantization mode discrimination processing section 7 selects one quantization characteristic expected to be most proper for quantization for the prediction error signal caused in the block from the dynamic range among plural quantization characteristics possessed by the quantization processing section 2. The signal quantized and coded by the quantization processing section 2 and the signal representing the quantization mode specifying the quantization characteristic by the quantization mode discrimination processing section 7 are outputted as a coded signal and a mode signal respectively. Thus, the coding with less picture quality deterioration is attained.

Description

DETAILED DESCRIPTION OF THE INVENTION [Field of Industrial Application] The present invention relates to an image signal encoding system and apparatus using predictive encoding that is resistant to transmission path errors 9 and highly efficient.

[Conventional technology]

An example of conventional predictive coding is shown in FIG. The input image signal is subtracted from the predicted signal by a subtracting means 45 to obtain a predicted error signal. The prediction error signal is quantized by quantization means 46. As an example, in the case of 5-bit equal-length encoding, the data will be quantized into 32 levels of values. Generally, the prediction error signal is generated concentrated around 0 as shown in Figure 10, so if the prediction error is close to 0 as shown in Figure 3, it is finely quantized and the prediction error is reduced. If it is far from O, it is coarsely quantized. The signal quantized by the quantization means 46 is output as a coded signal, and is also input to the inverse quantization stage 47. In the inverse quantization stage 47, the quantized level value is converted into a representative value, which is input to the addition process 48. In the addition process 48, the signal is added to the predicted signal to become a locally decoded signal. The locally decoded signal is input to the prediction means 49, subjected to prediction calculation and delay, and fed back to the subtraction means 45 and addition means 48. What is delayed by one sample in the prediction means 49 is called a previous value prediction, and what is delayed by one line is called a previous line prediction.
Prediction that is delayed by one frame is called interframe prediction. 1.2, which uses the previous value and the average value of the previous line as a prediction signal.
There is also dimensional prediction.

[Problem to be solved by the invention]

However, conventional predictive coding has the disadvantage that if a bit error occurs in the output encoded signal on the transmission path, the error spreads to the data that is subsequently restored on the decoding side, degrading the image quality. was there. Also, as an example, 7 and 5
When performing bit equal length encoding, if a large prediction error occurs at an edge, etc., as shown in Figure 3, all prediction errors that exceed the dynamic range are converted to the same representative number, resulting in a large quantization error. Unfortunately, image quality deteriorates due to gradient overload noise and the like. The present invention makes image quality deterioration less noticeable by confining the effect of bit errors in small blocks even if bit errors occur in the transmission path, and by changing the quantization characteristics on a block-by-block basis. It is an object of the present invention to provide an image signal encoding system that provides quantization characteristics according to the characteristics of edges, edges, flat areas, etc., and that causes less deterioration in image quality and an apparatus that has a simple configuration.

[Means to solve the problem]

The image signal encoding method of the present invention divides the screen into a plurality of blocks of a predetermined size, and each block has one
9- Calculate the prediction error between the value predicted from the pixel in the same block and the value of the original pixel, and prepare a plurality of quantization modes to specify the quantization characteristics when quantizing the prediction error. quantization characteristics are estimated from the dynamic range of prediction error for each block obtained from the original image signal, one of the plurality of quantization modes is selected for each block from the estimation result, and the selected quantization mode is The prediction error is quantized and encoded using the quantization characteristics of the quantization mode, and the encoded signal and a mode signal representing the selected quantization mode are output. The forward image signal encoding device (of the present invention) includes a quantizer selection control means for determining a dynamic range of a prediction error for each block from an original image signal and estimating a quantization characteristic from the dynamic range; a pixel position detection means for detecting the position of a pixel currently being encoded, a subtraction means for calculating a prediction error, a plurality of quantization means for quantizing the prediction error, and a plurality of quantization means for quantizing the prediction error. An image signal encoding device comprising a plurality of inverse quantization means, each having a characteristic opposite to the quantization characteristic of the quantization means, an addition means, a plurality of prediction means, and a plurality of selection means. .

The plurality of quantization means and the plurality of inverse quantization means each correspond to the plurality of prediction means, and each of the plurality of quantization means is a plurality of quantization means for quantizing and encoding with a plurality of quantization characteristics. It is possible to replace the quantization means with a plurality of inverse quantization means having characteristics opposite to the quantization characteristics of the plurality of quantization means.

Further, the quantizer selection control means includes a subtraction means for calculating a prediction error, a plurality of prediction means, a selection means,
Absolute value calculation means for calculating the absolute value of the prediction error; Area maximum value calculation means for calculating the maximum value of the absolute value of the prediction error of each prediction area corresponding to the plurality of prediction means; , a plurality of threshold determination means for classifying the quantization modes according to the maximum absolute value of the prediction error in each of the prediction regions; and a comprehensive determination means for determining the quantization mode of the quantization mode.

? [Operation] In the image signal encoding method of the present invention, the screen is divided into blocks of a predetermined appropriate size, one pixel in each block is outputted as one PCM signal, and the remaining picture j'
．． ri The difference between the value predicted from the pixel in the same block or the locally decoded signal and the original pixel value is quantized and encoded and output, but the quantization characteristics when quantizing and encoding are One of the respective quantization characteristics specified by the quantization mode of is selected for each block. The selection method is to find the dynamic range of the prediction error from the original image signal, and from that dynamic range, estimate the most suitable quantization characteristic among the custom quantization characteristics of the multiple quantization modes mentioned above, and the estimation result Choose from. Then, a quantized and encoded encoded signal and a mode signal representing the selected quantization mode are output.

〔Example〕

An embodiment of the image signal encoding method of the present invention will be described with reference to FIG. Looking at Figure 1, is it 1? In the measurement error calculation process 1, the error of the prediction signal output from the intra-block prediction process 5 for the original image signal, that is, the prediction error signal, is calculated. Then, quantization and encoding are performed in quantization processing 2, but the quantization characteristic during quantization is one of the quantization characteristics specified by the plurality of quantization modes. This is done by selecting one. As an example, assume that two types of characteristics shown in FIGS. 3 and 4 are prepared. On the other hand, prediction error dynamic range calculation process 6
Now, the dynamic range of the prediction error for each block, that is, the maximum amplitude of the prediction error signal occurring within that block, is determined from the original image signal. Then, in the quantization mode determination process 7, from the above-mentioned dynamic range, the quantization process 2 selects a plurality of quantization characteristics that are considered to be most suitable for quantizing the prediction error signal generated in that block. Judgment is made by selecting one of the characteristics. For example, in the quantization characteristics shown in Figure 3, the dynamic range is DA
So, in the quantization characteristics shown in Figure 4, the dynamic range is DB, so DA<DB? Suppose there is a relationship. In other words, the quantization characteristic in Figure 3 has a fine quantization characteristic, so the dynamic range is small;
@． That means no. Therefore, the prediction error dyna
If the dynamic range calculated in the range calculation process 6 is smaller than A in Fig. 3, then the quantization characteristic in Fig. 3, which has finer quantization, is more suitable9, than A in Fig. 3. If it is large, the quantization characteristic shown in FIG. 4, which has a wide dynamic range although the quantization becomes coarse, is more suitable. In quantization processing 2, quantization and encoding are performed using the appropriate quantization characteristic. The signal quantized and coded in quantization process 2 and the signal representing the quantization mode specifying the quantization characteristic selected in quantization mode determination process 7 are output as a coded signal and a mode signal, respectively. Ru.

! f. In addition, in the dequantization process 3, the inverse quantization characteristic of the quantization mode selected in the quantization process 2 is used to convert into a representative value. Then, the representative value and intra-block prediction processing 5
Local decoding is performed in local decoding processing 4 using the predicted signal output from , intra-block prediction is performed on the locally decoded signal in intra-block prediction processing 5 , and the predicted signal is subjected to prediction error calculation processing 1 and local Used in decoding process 4. An example of a prediction method in the intra-block prediction process 5 will be explained using FIG. 2. In FIG. 2, the screen is divided into a plurality of blocks of 4 pixels x 4 lines, and one of the blocks is described.

In the encoding of pixels in area 202 in the block shown in FIG. 2, the original pixels are output as they are, so no prediction is performed. In encoding the pixels included in the area 205, prediction is performed on the previous line, and in encoding pixels included in the area 205, prediction is performed using the previous value and the average value of the previous line. In the image signal encoding method of the present invention described above, predictive coding can be completed within a small block, and quantization characteristics can be adaptively selected for each block, so high-quality decoded images can be obtained. can be obtained.

Next, an image signal encoding device according to a second aspect of the present invention will be explained using FIG. 5.

In FIG. 5, the input image signal is input to a quantizer selection control means 11, a subtraction process 12, and a selection process 17. On the other hand, the pixel position detection stage 28 detects the position of the pixel currently being encoded, and sends a signal representing the pixel position information to the quantizer selection control stage]1, selection means 17, 22.2
Enter 7. The subtraction means] 2 calculates a prediction error from the input image signal and the prediction signal output from the selection process 27, and inputs it to the quantization means 13, 14, . . . , 15. The quantization means 13, 1.4, . . . , 15 each have a quantization characteristic specified by predetermined quantization modes 1 to n, and perform quantization and encoding processing, respectively, according to the quantization characteristic. are input into the selection means 16. The quantizer selection control means 11 calculates the Guinami range of prediction error for each block from the input image signal and the pixel position information output from the pixel position detection means 28, and calculates the prediction error generated in that block from the dynamic range. The quantization means 13, 14 . ,...
, 15 is selected, and a signal representing a quantization mode specifying the selected quantization characteristic is output. The output of the quantizer selection control means 11 is output as a mode signal, and is also outputted to the selection means 16 . 21
is input.

The selection means 16 selects the quantized and encoded signal of the quantization mode indicated by the mode signal, and selects the quantized and encoded signal in the selection stage 17.
Enter. In the selection means 17, the pixel position detection means 28
According to the pixel position information output from the pixel position information output from the pixel position information, if the position of the pixel currently being encoded is included in the area 202 shown in FIG.
4, if it is included in either of 205, select means 16
Select the output of The output of the 1,000 selection stages 17 is output as an encoded signal, and is also input to the 1,000 selection stages 22 and the dequantization means 18, 19, . . . 20. The dequantization means 18, 19, . Convert to value.

The outputs of the inverse quantization means 18, 19, . . . , 20 are input to a selection stage 21, respectively. The selection stage 21 selects the dequantized signal of the quantization mode indicated by the output signal of the quantizer selection control means 11, and inputs it to the selection stage 22. In the selection means 22, in accordance with the pixel position information outputted from the pixel position detection means 28 in the same manner as the selection step 17, if the position of the pixel currently being encoded is included in the area 202 shown in FIG. selecting the output of means 17;
Areas 203, 204. , 205, the output of the selection means 21 is selected. The output of the selection stage 22 is input to the addition means 23. The addition means 23 adds the output of the selection means 22 and the predicted signal of the output of the selection means 27, and outputs the result as a locally decoded signal to the prediction means 24, 25.2.
6 is input. Prediction means 24, 25. 26, for example, the prediction means 24 performs previous value prediction, and the prediction means 2
5 performs previous line prediction, and prediction means 26 performs prediction based on the previous value and the average value of the previous line. Prediction means 24, 25.2
The outputs of 6 are respectively input to the selection means 27. A value of "0" is also input to the selection stage 27, and according to the output of the pixel position detection means 28, if the position of the pixel currently being encoded is included in the area 202 in FIG. Select a value, and if it is included in the area 203, select the output of 1,000 predicted steps 24, if it is included in the area 204, select the output of 1,000 predicted steps 25, and if it is included in the area 205, select the output of 1,000 predicted steps 26.
Select the output of The output of the selection stage 27 is input to the subtraction stage 12 and the addition means 23 as a prediction signal. By configuring the image signal encoding device described above, the above-described image signal encoding method can be realized.

1. The plurality of quantization means and inverse quantization means applied to the image signal encoding device described below correspond to the plurality of prediction methods, and each prediction method has a plurality of quantization modes. Image signal encoding according to the third invention, in which the quality of a decoded image can be further improved by using a quantization means that performs quantization and encoding using the quantization means, and an inverse quantization means corresponding to the quantization means, respectively. The apparatus will be explained below with reference to FIG.

In FIG. 6, the quantizer selection control means 51 and the ten subtraction stages 5
2 and the pixel position detection means 95 are the same as those of the quantizer selection control stage 11, the subtraction means 12, and the pixel position detection stage 28 shown in FIG. 5, so a description thereof will be omitted here. . The prediction error signal output from the thousand subtraction stages 52 is quantized by quantization means 53 , 54 , . . . , 55 ,
56, 57,..., 58, 59, 60,...
. , 61, respectively. Each quantization means corresponds to one of the prediction methods of a plurality of prediction means described later, and the quantization means 53, 54 . ,..., 55 is the prediction method of the prediction means 91, and the quantization means 56 1 57 H
, 58 correspond to the prediction method possessed by the prediction means 92, and the quantization means 59, 60, . . . , 61 correspond to the prediction method possessed by the prediction means 93. 1, quantization means 53, 56.
59 has a quantization characteristic of quantization mode 1, and the quantization means 54, 57 . 60 has quantization characteristics of quantization mode 2, and quantization means 55, 58. 61 has a quantization characteristic of quantization mode n. Each quantization means quantizes and encodes according to its own quantization characteristics, and the outputs of the quantization means 53, 54, . . . , 55 are input to the selection means 62, and the quantization means 56, 57, .・,5
The output of 8 is input to the selection means 63, and the quantization means 59,
The outputs of 60, . . . , 61 are input to a selection stage 64. Selection means 62, 63. At 64, signals quantized and encoded with the quantization characteristics specified by the corresponding quantization mode are selected according to the mode signal output from the quantizer selection control means 51, and each output is input to the selection means 65. Ru. In accordance with the output of the pixel position detection means 95, the selection means 65 selects the input image signal if the position of the pixel currently being encoded is included in the button area 202 in FIG. If so, the output of the selection means 62 is selected, if it is included in the area 204, the output of the selection means 63 is selected, and if it is included in the area 205, the output of the selection means 64 is selected. The output of the selection means 65 is outputted as an encoded signal, and is also sent to the selection means 78 and the dequantization means 66, 67, . . . , 68, 69, 70, .
・,71. , 72, 73, . . . , 74, respectively. Inverse quantization means 66, 67,..., 68
,69,70,...,71,72,73,...
, 74 are quantization means 53, 54, . . . 5, respectively.
5, 56, 57,..., 58, 59, 60,...,
It has an inverse characteristic corresponding to the quantization characteristic of 61. Each inverse quantization means converts the input level signal into a representative value, and the outputs of the inverse quantization means 66, 67, . The output of 71 is sent to the selection means 76, and the inverse quantization means 72, 73, . . .
, 74 are input to selection means 77, respectively. Selection 1,000 steps 75, 76. 77, the selection means 62, 63
．． Similarly to 64, according to the mode signal output from the quantizer selection control means 51, signals converted to representative values with the inverse quantization characteristic specified by the corresponding quantization mode are selected, and the signals are sent to the selection means 78. is input. Selection means 78
Now, in the same way as the selection means 65, the image position detection means 9
According to the output of 5, if the position of the pixel currently being encoded is included in the area 202 shown in FIG. If it is selected and included in the area 204, the selection means 76
Select the output of , and if it is included in the area 205, select 1,000 steps 7
Select output 7. The output of the selection means 78 is added to the addition means 7
9 is input. Hereinafter, the addition means 79, the prediction means 91,
92, 93, and the configuration and operation of the selection means 94.
The addition stage 23, prediction means 24, 25,
26 and 1,000 selection steps 27, so the explanation will be omitted here. By configuring the image signal encoding device of the third invention described above, the aforementioned image signal encoding method can be realized, and processing can be performed with quantization characteristics suitable for each prediction method. Therefore, it is possible to further improve the quality of the decoded image.

An image signal encoding apparatus according to a fourth invention in which the quantizer selection control means in the image signal encoding apparatuses according to the first, second, and third inventions is specifically constructed will be described below. 5 and 6, quantizer selection control means 1. ]. ,51
The configuration and operation of the other parts are the same as those described above, so the explanation will be omitted here. The configuration of the quantizer selection control means 11.51 is shown in FIG. The input image signal in FIG. 7 is input to subtraction means 105 and prediction means 101, 102, 1.03. Prediction means 10
1, 1.02, and 103 are the prediction means 24, 25.26 in FIG. 5, and the prediction means 91, 92. in FIG. 6, respectively. It is the same as 93. That is, for example, the 1,000-stage prediction 101 performs previous value prediction, and the 1,000-stage prediction 1
0 2 predicts the previous line, and the prediction stage 103 predicts the previous value and the average value of the previous line.

The prediction means ion, 102, according to a predetermined method.
The signals predicted at 103 are each input to selection means 104. The selection means 104 is the same as the selection stage 27 in FIG. 5 and the selection stage 94 buttoned in FIG. 6, and performs the same operation. That is, if the position of the pixel of the input image signal is included in the area 202 shown in FIG.
If the value is included in the area 203, the prediction means 10
] is included in the area 204, the output of the prediction means 102 is selected, and if it is included in the area 205, the output of the prediction means 103 is selected.

The output of the selection means 104 is subtracted by the subtraction means 1. 05, and the difference with the input image signal is taken to obtain a prediction error that does not include quantization processing. Subtraction means]. The prediction error outputted from 05 is input to the absolute value calculation means 106 to calculate the absolute value, and the absolute value calculation means 1. The outputs of 06 are input to area maximum value calculation means 107, 1.08, and 109, respectively. Area maximum value calculation means 107, 1.0
8.1.09 calculates the maximum absolute value of the prediction error, that is, the dynamic range, for each region of each prediction method. Specifically, the region maximum value calculation means 107 calculates the dynamic range of the prediction error in the region 203 (previous value prediction region) shown in FIG.
The dynamic range of the prediction error in the (previous line prediction area) is determined, and the area maximum value calculation means 109 determines the dynamic range of the prediction error in the area 205 (previous value and average value prediction area of the previous line). Area maximum value calculation means 1.07,
The dynamic ranges of the prediction errors obtained in steps 108 and 109 are determined by threshold determination means 110 and 11.
1, 1, and 12, respectively.

The operation of the threshold value determination means 110, 1.11, and 112 will be explained using FIG. In Fig. 8, the input dynamic range is O to threshold 1 (hereinafter 11
] 1), quantization mode 1 specifies a quantization characteristic with fine quantization and a dynamic range of th1.
, when the dynamic range is thl~t1]2,
The quantization is based on the characteristics of quantization mode 1. Quantization mode 2 specifies the quantization characteristics where 9 is coarse but the dynamic range is th2. Select quantization mode from Namino Cleanse.

Threshold determination means 110,]. 1]. , 112, determinations are made using different threshold values, and a threshold determination suitable for each prediction method is performed. Each threshold judgment result is input to the comprehensive judgment stage 1]3, and for example, the judgment result in the threshold judgment means 11.0 and 1.11 is mode 1,
If the judgment result in the 1,000-stage threshold judgment 112 is mode 2, then the judgment result of the 1,000-stage threshold judgment 112, in which the judgment by the Daiichi check range is the largest, that is, mode 2, is pressed in that block. A comprehensive judgment is made within the block, and the result is output as a mode signal for each block. By configuring the quantizer selection control means described above, it is possible to configure an image signal encoding device according to the fourth aspect of the invention, which realizes the above-described image signal encoding method.

However, in this example, the block size is 4 pixels x 4.
Although it has been described as a line, larger or smaller blocks may also be used. Regarding the prediction method,
Here, an example in which previous value prediction, previous line prediction, and average value prediction of previous value and previous line are combined has been described, but other combinations may be used as long as prediction processing can be completed within a block. 1. In the comprehensive judgment means in Fig. 7,
Here, a method of selecting the mode with the largest determination by Dynamino Cleanse was used, but other methods may be used.

〔Effect of the invention〕

As explained above, according to the present invention, predictive coding is completed within a block, so even if a bit error occurs due to a button on the transmission path, the effect is confined to a small block, and image quality deterioration is not noticeable. 1. Since it is possible to give each block a quantization characteristic suitable for characteristics such as image fineness and flatness, it is possible to obtain a high-quality decoded image. Since the quantization mode judgment for each image is estimated from the original image signal, it is only necessary to perform the quantization mode selection process using the judgment result during encoding processing, which simplifies the hardware configuration.
It is possible to provide an image signal encoding method and device that are extremely useful in practice.

[Brief explanation of drawings]

FIG. 1 is a diagram showing the flow of the image signal encoding method of the present invention,
Figure 2 is an explanatory diagram of intra-block predictive coding, Figures 3 and 4.
The figure is an explanatory diagram of an example of quantization characteristics, FIG. 5 is a configuration diagram of an image signal encoding device of the second invention, FIG. 6 is a configuration diagram of an image signal encoding device of the third invention, and FIG. 8 is an explanatory diagram of threshold value determination, FIG. 9 is a diagram of conventional predictive coding, and FIG. 10 is an explanatory diagram of prediction error occurrence distribution. In the figure, 1 is a prediction error calculation process, 2 is a quantization process, 3 is an inverse quantization process, 4 is a local decoding process, 5 is an intra-block prediction process, 6 is a prediction error dynamic range calculation process,
7 is a quantization mode determination process, 11.51 is a quantizer selection control means, 28.95 is a pixel position detection means, 12, 52
, 45, 105 are subtraction means, 23, 79.48 are addition means, 13, 14. ,1.5,53,54,55,56,
57, 58, 59, 60, 6]. , 46 is a quantization means,
18, 19, 20, 66, 67, 68, 69, 70, 7
1, 72, 73. 74-, 47 are inverse quantization means; 24.
, 25, 26, 91. ,92,93,49,1.01,
102, 103 are prediction means, 16, 17, 21, 22,
27, 62, 63, 64, 65, 75, 76, 77, 7
8, 94, 104 are selection means, 106 is an absolute value calculation means, 107, 108, 109 are area maximum value calculation means,
11.0, 111, 112 are threshold determination means, 113
is a comprehensive judgment means, 201 is a block, 202, 203,
204. , 205 is the area of the same prediction method, 81.82 is the quantization characteristic, 83 is the prediction error distribution, and 84 is the dynamic range axis of the prediction error.

Claims (4)

[Claims] (1) Divide the screen into multiple blocks of a predetermined size, calculate the prediction error between the value predicted from the pixels in the same block and the value of the original pixel in each block, and calculate the prediction error. Prepare multiple quantization modes that specify the quantization characteristics during quantization, estimate the quantization characteristics from the dynamic range of the prediction error for each block obtained from the original image signal, and use the estimation results to determine the quantization characteristics. quantization mode is selected for each block, the prediction error is quantized and encoded using the quantization characteristics of the selected quantization mode, and the encoded signal and the selected quantization mode are quantized and encoded. An image signal encoding method characterized by outputting a mode signal representing a quantization mode. (2) Quantizer selection control means for determining the dynamic range of prediction error for each block from the original image signal and estimating quantization specificity from the dynamic range, and for detecting the position of the pixel currently being encoded. a pixel position detection means, a subtraction means for calculating a prediction error, a plurality of quantization means for quantizing the prediction error, and an inverse characteristic of the quantization characteristic of the plurality of quantization means, respectively. a plurality of inverse quantization means, an addition means, a plurality of prediction means,
An image signal encoding device comprising a plurality of selection means. (3) A plurality of quantizations for quantizing and encoding the plurality of quantization means and the plurality of inverse quantization means respectively corresponding to the plurality of prediction means and each with a plurality of quantization characteristics. 3. The image signal encoding apparatus according to claim 2, wherein said means is replaced with a plurality of inverse quantization means having a characteristic opposite to the quantization characteristic of said plurality of quantization means. (4) The quantizer selection control means includes a subtraction means for calculating a prediction error, a plurality of prediction means, and a selection means,
Absolute value calculation means for calculating the absolute value of the prediction error; Area maximum value calculation means for calculating the maximum value of the absolute value of the prediction error of each prediction area corresponding to the plurality of prediction means; , a plurality of threshold determination means for classifying the quantization modes according to the maximum absolute value of the prediction error in each of the prediction regions; 3. The image signal encoding apparatus according to claim 2, further comprising comprehensive determination means for determining the quantization mode of the image signal.