JPH04339472A - Image signal forecasting coder/decoder and smoothing device - Google Patents

Abstract

PURPOSE:To hold the average density of a unit area in a decoded image and to suppress granular noises. CONSTITUTION:A coder adds a feedback quantizing error signal 46 and an input image signal to each other and obtains a corrected image signal 32. A quantizing circuit 35 quantizes a differential signal 34 between the signal 32 and a forecasting signal 41. A representative value forming circuit 37 reversely quantizes a quantized signal 36. A forecasting circuit 40 forecasts an image signal to be coded next by a preceding local decoded signal 39 and outputs a forecasted image signal 41. An adder 38 adds the signal 41 to a representative value signal 47. The signal 39 is subtracted from the signal 32 to obtain a quantized error signal 44. The signal 44 is passed through an one-picture element delay circuit 45 and fed back to the succeeding input image signal. A decoder is provided with a smoothing circuit for reducing the degree of smoothing of a decoded signal in the ascending order of the edge component of the decoded signal.

Description

[Detailed description of the invention]

0001

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a picture signal predictive encoding/decoding device for compressing and expanding a picture signal, and a picture signal smoothing device for removing noise from a picture signal containing granular noise.

0002

BACKGROUND OF THE INVENTION In recent years, image signal predictive encoding and decoding devices (so-called DPCM) can be configured with smaller hardware than encoding and decoding devices that use orthogonal transform, and are lower in cost and operating speed. It is expected to be applied to digital electronic still cameras, etc.

An example of the above-mentioned conventional image signal predictive encoding device and decoding device will be described below with reference to the drawings.

FIG. 8 shows a block diagram of a conventional image signal predictive coding device. In FIG. 8, a subtracter 2 subtracts a predicted image signal 13 from an input image signal 1 and outputs a difference signal 14. The quantization circuit 3 nonlinearly quantizes the difference signal 14 and outputs a quantized signal 15. Local decoding circuit 17
is composed of a representative value generation circuit 4, an adder 5, a limiter 8, and a prediction circuit 6. The representative value generation circuit 4 dequantizes the quantized signal 15 and outputs a representative value signal 16. The adder 5 adds the predicted image signal 13 output from the prediction circuit 6 and the representative value signal 16. Limiter 8 limits the output signal of adder 5 to the same dynamic range as input image signal 1, and outputs locally decoded signal 18. The prediction circuit 6 predicts the image signal to be encoded next using the previous locally decoded signal 18 and outputs a predicted image signal 13. 7 is a variable length encoding circuit such as Huffman encoding, which assigns a shorter code to a value with a higher probability of occurrence of the quantized signal 15, and outputs an encoded image signal 19.

The operation of the image signal predictive encoding device configured as described above will be explained.

Since the input image signal 1 and predicted image signal 13 are signed 8-bit data, the difference signal 14 is signed 9-bit data and takes a value from -255 to 255. (Table 1) shows the relationship between input and output data of the quantization circuit 3.

[0007]

[Table 1]

[0008] The quantized signal 15 is expressed using signed 3 bits to reduce the amount of data. Usually, due to the correlation of image signals,
The closer the values of the difference signal 14 and the quantized signal 15 are to 0, the more frequently they occur. Therefore, by variable length encoding the quantized signal 15, the amount of data can be further reduced.

The relationship between input and output data of the representative value generation circuit 4 is shown in Table 2.

0010

[Table 2]

FIG. 9 shows a block diagram of the prediction circuit 6 when using pre-prediction with a prediction coefficient of 1. In FIG. 9, 20
delays the local decoded signal 18 by one pixel and outputs it as the predicted image signal 13.

FIG. 10 shows a block diagram of a conventional image signal predictive decoding device. The variable length code decoding circuit 21 decodes the variable length coded quantized difference signal 19. The representative value generation circuit 22 dequantizes the quantized difference signal and outputs a representative value signal 23. The adder 24 adds the representative value signal 23 and the predicted image signal 27. The limiter 29 limits the output signal of the adder 24 to the same dynamic range as the input image signal 1, and outputs a decoded signal 25.

The prediction circuit 26 predicts the next image signal using the previous decoded signal 25 and outputs a predicted image signal 27. The portion surrounded by the dotted line in FIG. 10 has the same configuration as the local decoder in FIG. 8.

(For example, "Multidimensional signal processing of TV images" 2
(pages 21-223).

[0015]

However, in the above configuration, if the quantization step of nonlinear quantization is coarsened in order to increase the compression rate of image data, the quantization error increases. At this time, in the decoded image signal, the average density per area is non-conservative due to quantization error, the quantization error is fed back to the next pixel, creating a texture that is correlated in the scanning direction, and the increase in granular noise. There was a problem of image deterioration due to

In view of the above problems, the present invention provides an image signal predictive encoding device and a decoding device that can obtain a good decoded image even if the quantization step is coarsened in order to increase the compression rate of image data. It is something to do.

[0017]

[Means for Solving the Problems] In order to solve the above-mentioned problems, the image signal predictive coding device of the present invention uses a prediction method that predicts a decoded image signal and an image signal to be encoded next from an encoded image signal. local decoding means for outputting an image signal; quantization means for quantizing the difference between the decoded image signal from the local decoding means and the image signal to be encoded; and a quantization means for quantizing the difference between the decoded image signal and the image signal before encoding. The encoder includes an encoding error feedback means for feeding back the image signal to be encoded.

Further, the image signal decoding device of the present invention includes local decoding means for decoding the encoded image signal, and quantization means for quantizing the difference between the decoded image signal from the local decoding means and the input image signal. A decoding device that decodes a signal encoded by an image signal predictive encoding device includes a decoding means for decoding the encoded image signal, and a smoothing means for smoothing the decoded image signal.

Furthermore, the smoothing means includes an edge component detection means for outputting an edge component signal of the image signal, and changes the degree of smoothing depending on the magnitude of the edge component signal.

[0020]

[Operation] With the above-described configuration, the present invention preserves the average density per area by feeding back the quantization error to the image signal to be encoded, suppresses the occurrence of texture by distributing the quantization error, and smooths the image signal. The grading means can reduce granular noise and prevent important edge information from being degraded in image information.

[0021]

DESCRIPTION OF THE PREFERRED EMBODIMENTS A picture signal predictive encoding device and decoding device according to an embodiment of the present invention will be described below with reference to the drawings.

FIG. 1 is a block diagram of an embodiment of an image signal predictive coding apparatus according to the present invention. The adder 31 adds the feedback quantization error signal 46 and the input image signal 30 and outputs an error-corrected image signal 32. The subtracter 33 subtracts the predicted image signal 41 from the error corrected image signal 32 and outputs a difference signal 34.

The quantization circuit 35 nonlinearly quantizes the difference signal 34 and outputs a quantized signal 36. The local decoding circuit 42 includes a representative value generation circuit 37, an adder 38, a limiter 51, and a prediction circuit 40. The representative value generation circuit 37 dequantizes the quantized signal 36 and outputs a representative value signal 47. The adder 38 adds the predicted image signal 41 output from the prediction circuit 40 and the representative value signal 47.

The limiter 51 limits the output signal of the adder 38 to the same dynamic range as the input image signal 30 and outputs a locally decoded signal 39. The prediction circuit 40 predicts the image signal to be encoded next using the previous locally decoded signal 39 and outputs a predicted image signal 41.

The subtracter 43 subtracts the local decoded signal 39 from the error corrected image signal 32 and outputs a quantization error signal 44. The one-pixel delay circuit 45 delays the quantization error signal 44 by one pixel in order to feed it back to the next input image signal.
A feedback quantization error signal 46 is output. The variable length encoding circuit 48 assigns a shorter code to a value with a higher probability of occurrence of the quantized signal 36, and outputs an encoded image signal 49.

Table 1 shows the relationship between the input and output data of the quantization circuit 35 when the input image signal 30 is signed 8-bit data.

The quantized signal 35 is expressed using signed 3 bits to reduce the amount of data. Usually, due to the correlation of image signals,
The closer the values of the difference signal 34 and the quantized signal 36 are to 0, the more frequently they occur. Therefore, by variable length encoding the quantized signal 36, the amount of data can be further reduced.

Table 2 shows the relationship between the input and output data of the representative value generation circuit 37. As described above, according to this embodiment, by feeding back the quantization error (feedback quantization error signal 46) of the decoded image signal to the image signal to be encoded, the quantization error can be reduced in the image to be encoded thereafter. Since it is corrected in the signal, the average concentration per area can be preserved.

FIG. 2 is a block diagram of another embodiment of the portion 52 in FIG. The error distribution means 60 distributes the quantization error signal 44 to subsequent input image signals, and also distributes a feedback quantization error signal to be added to the current input image signal from among the quantization errors of previously encoded image signals. 4
Outputs 6.

FIG. 3 is an explanatory diagram of the operation of the error distribution means 60 of FIG. 2, and the operation will be explained below.

The quantization error Et generated when pixel X is encoded is distributed to pixel A, which is encoded next to pixel X, and four pixels B, C, and D, which are encoded in the next scan. . The distributed quantization errors at pixel X distributed to pixels A, B, C, and D are Ea, Eb, Ec, and Ed, respectively.

As shown in FIG. 3, there are two groups of error distribution coefficients for distributing the quantization error Et to the four pixels: Ka and Kb. As the error distribution coefficient, either Ka or Kb is selected by the distribution coefficient selection signal S. The distribution coefficient selection signal S is output from a random signal generation circuit R that generates a random signal for each pixel to be encoded. Distributed quantization errors Eb, Ec distributed to pixels B, C and D,
Since Ed is used for encoding in the next scan, it is stored in the line memory. Distribution quantization error Eb of line memory,
The distributed quantization error when encoding the pixel immediately before X is already stored in the location where Ec is stored. Therefore,
When storing the distributed quantization errors Eb and Ec in the line memory, the already stored distributed quantization errors are added and then stored in the line memory.

The feedback quantization error signal 46 in FIG. 2 is generated when the immediately preceding pixel of the input image signal is encoded.
It is generated by adding the distributed quantization error distributed to A and the distributed quantization stored in the line memory when the input image signal of the immediately previous scan was encoded.

As described above, according to this embodiment, since the average density per area can be preserved by providing the feedback means for the image signal that encodes the quantization error, the image quality of the decoded image is improved. Furthermore, the feedback means
By providing error distribution means for distributing quantization errors to neighboring pixels of the encoded pixel, the quantization errors do not significantly affect the next pixel, and image quality can be improved. In addition, by randomly switching a plurality of error distribution coefficient groups for each pixel, texture caused by quantization error distribution that occurs in a decoded image can be suppressed.

FIG. 4 is a block diagram of an embodiment of an image signal decoding device according to the present invention. The encoded image signal 49 is shown in FIG.
This is a signal encoded by an image signal encoding device. The variable length code decoding circuit 70 returns the encoded image signal 49 subjected to variable length encoding to the signal before variable length encoding. The variable length code decoding circuit 70 outputs a quantized signal 71. The representative value generation circuit 72 dequantizes the quantized signal 71 and outputs a difference signal 73. The relationship between the input and output data of the representative value generation circuit 72 is expressed as (
Table 2) shows the results.

Adder 74 adds predicted image signal 77 and difference signal 73. The limiter 82 limits the output signal 75 of the adder 74 to the dynamic range of the image signal to be decoded, and outputs a decoded signal 83. The prediction circuit 76 is shown in FIG.
The prediction circuit 40 predicts the next image signal to be decoded using the previously decoded signal 83 and generates the predicted image signal 77.
Output.

The edge component detection circuit 80 detects the edge component of the decoded signal 83 and outputs an edge component signal 81. The smoothing circuit 78 smoothes the decoded signal 83 according to the edge component signal 81 and outputs a decoded image signal 79. The smoothing circuit 78 reduces the degree of smoothing as the edge component signal 81 increases.

FIG. 5 is an explanatory diagram of the operation of the edge component detection circuit 80 in FIG. 4. The decoded signal 83 is scanned in a 3 x 3 pixel window and the absolute value of {(decoded signal at the center of the window) - (sum of the decoded signals at the four corners of the window)/4} is calculated. Calculate edge components of decoded pixels. That is, the second-order differential component of the decoded signal within the window is calculated, and the absolute value of this second-order differential component is output as the edge component signal 81.

As another embodiment of the edge component detection circuit 80 in FIG. 4, the difference between the maximum value and the minimum value of the decoded signal within the 3×3 window may be output as the edge component signal 81.

FIG. 6 is a block diagram of a first embodiment of smoothing circuit 78 in FIG. The average value circuit 90 scans the decoded signal 83 in a 3×3 pixel window, calculates the average value of the decoded signals of 9 pixels within the window, and outputs the average value signal 91 of the decoded signals at the center of the window. The one-line delay circuit 96 delays the decoded signal 83 by one line scan and outputs a line-delayed decoded signal 97. Comparator 93
compares the edge component signal 81 with a predetermined threshold value 95, and outputs a selection signal 94 that becomes high level when the edge component signal 81 is larger than the predetermined threshold value 95.

The selector 92 receives the average value signal 91 and the line delay decoded signal 97 as input. The selector 92 selects the line-delayed decoded signal 97 when the selection signal 94 is at a high level, and outputs the decoded image signal 79.

As described above, the smoothing circuit 78 does not perform smoothing when the edge component signal 81 is greater than the threshold value 95, but performs smoothing processing only when the edge component signal 81 is less than or equal to the threshold value 95.

FIG. 7 is a block diagram of a second embodiment of the smoothing circuit 78 in FIG. The average value circuit 90 scans the decoded signal 83 in a 3×3 pixel window, calculates the average value of the decoded signals of 9 pixels within the window, and outputs the average value signal 91 of the decoded signals at the center of the window.

The 1-line delay circuit 96 converts the decoded signal 83 into 1
A line-delayed decoded signal 97 is output after being delayed by the amount of line scanning. Multipliers 102, 103, 104, 105, 106, and 107 multiply the average value signal 91 or line delay decoded signal 97 by the coefficients shown in FIG. The multiplier that multiplies the coefficients is realized by a bit shift and an adder.

Adder 110 includes multipliers 102 and 10
The outputs of 3 are added and a signal 115 is output. Adder 11
1 adds the outputs of multipliers 104 and 105 and outputs signal 116. Adder 112 adds the outputs of multipliers 106 and 107 and outputs signal 117. A look-up table (hereinafter referred to as LUT) 114 receives the edge component signal 81 and outputs a selection signal 115. L
The relationship between input and output data of the UT 114 is shown in (Table 3).

[0046]

[Table 3]

Selector 113 receives average value signal 91, signal 115, signal 116, signal 117, and line delay decoded signal 97 as input. Selector 113 receives selection signal 1
One of the input signals is selected based on the value of 15, and a decoded image signal 79 is output. Selection signal 1 in selector 113
The relationship between the value of 15 and the input signal selected as the output is expressed as (
Table 4) shows the results.

[0048]

[Table 4]

By configuring the smoothing circuit 78 as described above, the degree of smoothing can be made smaller as the edge component becomes larger.

[0050] The image signal decoding device having the above configuration is as follows:
Even if the minimum quantization step of the quantizer of the encoding device is increased, granular noise occurring in the decoded image signal can be suppressed, and edge information can be reproduced without blurring.

In addition, if the device shown in FIG. 1 is used as the image signal predictive encoding device and the device shown in FIG. 4 is used as the image signal decoding device, the average density per area is saved at the time of encoding. In the embodiments of FIGS. 1 and 4, the effect of the smoothing process in the decoding device is further improved.
Although the prediction circuits 40 and 76 have been described using pre-prediction, two-dimensional prediction may also be used.

Furthermore, if an image signal smoothing device is configured using the smoothing circuit 78 and the edge component detection circuit 80 in FIG. 4, noise in the image signal including granular noise can be removed, and edge information in the image signal can be It is possible to realize an image signal smoothing device that does not cause deterioration of the image signal.

[0053]

As described above, the present invention provides an image signal predictive coding device with a quantization error feedback means for feeding back quantization errors to the input image signal, thereby making it possible to preserve the average density per area and improve decoding. The image quality of the image signal can be improved. Furthermore, when feeding back the quantization error, by providing an error distribution means for distributing the quantization error to a plurality of input image signals, the occurrence of texture can be suppressed and the image quality of the decoded image signal can be improved. Furthermore, by providing the image signal decoding device with a smoothing means that can change the amount of smoothing according to the size of the edge component of the image signal, it is possible to reduce grainy noise that occurs during encoding, and to can prevent important edge information from being degraded. Furthermore, by using the smoothing means for image signals containing granular noise, it is possible to realize an image signal smoothing device having similar effects.

[Brief explanation of drawings]

FIG. 1 is a block diagram of an image signal predictive coding device in an embodiment of the present invention.

FIG. 2 is a block diagram of another embodiment of the portion 52 in FIG. 1;

FIG. 3 is an explanatory diagram of the operation of the error distribution means 60 in FIG. 2;

[Figure 4]
Block diagram of an image signal decoding device in an embodiment of the present invention

FIG. 5 is an explanatory diagram of the operation of the edge component detection circuit 80 in FIG. 4;

[
FIG. 6 is a block diagram of a first embodiment of the smoothing circuit 78 in FIG.

FIG. 7 is a block diagram of a second embodiment of the smoothing circuit 78 in FIG. 4;

[Fig. 8] Block diagram of a conventional image signal predictive coding device

[
FIG. 9 is a block diagram of the prediction circuit 6 when using pre-prediction with a prediction coefficient of 1 in FIG.

FIG. 10: Block diagram of a conventional image signal decoding device

[Explanation of symbols]

31 Adder 35 Quantization circuit 37 Representative value generation circuit 40 Prediction circuit 43 Subtractor 45 1 pixel delay circuit 60 Error distribution means 78 Smoothing circuit 80 Edge component detection circuit 90 Average value circuit 113 Selector

Claims (15)

[Claims] 1. Local decoding means for outputting, from an encoded image signal, at least a decoded image signal and a predicted image signal that predicts an image signal to be encoded next; quantization means for quantizing the difference and inputting the quantized image data to the local decoding means; and quantization for feeding back the error between the decoded image signal and the image signal before encoding to the image signal to be encoded thereafter. An image signal predictive coding device comprising error feedback means. 2. The local decoding means includes representative value generation means for inversely quantizing the quantized signal output from the quantization means, and predicting an image signal to be encoded next using a previously locally decoded decoded image signal. The image signal predictive encoding device according to claim 1, further comprising a prediction means for predicting. 3. The encoding error feedback means includes error distribution means for distributing the error between the decoded image signal and the unencoded image signal to the next encoded image signal and its neighboring pixels. image signal predictive coding device. 4. The error distribution means comprises a plurality of error distribution coefficients that determine the amount of error distributed to neighboring pixels, and an error distribution selection means that selects one of the plurality of error distribution coefficients. 3. The image signal predictive encoding device according to item 3. 5. The image signal predictive coding apparatus according to claim 4, wherein the error distribution selection means selects one of the error distribution coefficients for each image signal. 6. An image signal predictive encoding device comprising a local decoding means for decoding an encoded image signal, and a quantization means for quantizing a difference between a decoded image signal from the local decoding means and an input image signal. An image signal decoding device that decodes an encoded signal, the image signal decoding device comprising: decoding means for decoding the encoded image signal; and smoothing means for smoothing the decoded image signal. 7. The smoothing means comprises edge component detection means for outputting an edge component signal of the decoded image signal, and smoothing amount variable means for varying the degree of smoothing according to the magnitude of the edge component signal. Item 6. Image signal decoding device according to item 6. 8. The smoothing amount variable means includes a mixing means for generating a plurality of smoothed signals having different smoothing amounts by changing a mixing ratio of the smoothed image signal and the input image signal, and a smoothing amount output from the mixing means. 8. The image signal decoding apparatus according to claim 7, further comprising selecting means for selecting one of the plurality of smoothed signals according to the edge component signal. 9. The edge component detection means converts the decoded image signal into N
8. The image signal decoding device according to claim 7, wherein the image signal decoding device scans in a window of ×M pixels (N and M are natural numbers) and outputs the difference between the maximum value and the minimum value of the image signal within the window as an edge component signal. 10. The edge component detection means scans the decoded image signal in a window of N×M (N and M are natural numbers) pixels,
8. The image signal decoding device according to claim 7, wherein the second-order differential component of the image signal within the window is output as an edge component signal. 11. Local decoding means for outputting, from an encoded image signal, at least a decoded image signal and a predicted image signal that predicts an image signal to be encoded next; quantization means for quantizing the difference and inputting the quantized image data to the local decoding means; and quantization for feeding back the error between the decoded image signal and the image signal before encoding to the image signal to be encoded thereafter. An image signal predictive encoding/decoding device comprising an image signal predictive encoding device having error feedback means, and the image signal decoding device according to claim 6. 12. An image signal smoothing device for removing noise from an image signal, comprising: edge component detection means for detecting an edge component of an input image signal; An image signal smoothing device comprising a smoothing amount variable means that changes the degree of smoothing accordingly. 13. The edge component detection means scans the decoded image signal in a window of N×M (N and M are natural numbers) pixels,
The image signal smoothing device according to claim 12, wherein the difference between the maximum value and the minimum value of the image signal within the window is output as an edge component signal. 14. The edge component detection means scans the decoded image signal in a window of N×M (N and M are natural numbers) pixels,
The image signal smoothing device according to claim 12, wherein the second-order differential component of the image signal within the window is output as an edge component signal. 15. The smoothing amount variable means includes a mixing means for generating a plurality of smoothed signals having different smoothing amounts by changing a mixing ratio of the smoothed image signal and the input image signal, and a smoothing amount output from the mixing means. 13. The image signal smoothing device according to claim 12, further comprising selecting means for selecting one of the plurality of smoothed signals according to the edge component signal.