JPH05219496A - Picture encoder and moving picture encoder - Google Patents

Abstract

PURPOSE:To reduce the variation of the picture quality of each block. CONSTITUTION:A switch 122 is switched so that the quantization error of an input picture position applied from an inverse discrete cosine converter 22 can be inputted to an absolute value sum calculator 123 only when the input picture element is selected by a selector 121. The absolute value sum calculator 123 calculates the absolute value sum of the selected quantization error by each block unit, the calculated absolute value sum is compared with a set threshold value by a comparator 124, and a judgement signal indicating the result is outputted to a quantizer 16. When the judgement signal indicates that the absolute value sum is larger than the threshold value, the quantizer 16 quantizes a conversion factor stored in a memory 14 again by using a new quantization scale prepared by subtracting a value DELTA (>0) from the quantization scale.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a picture coding apparatus and a moving picture coding apparatus, and more particularly to an adaptive quantization technique in a coding apparatus using orthogonal transform.

[0002]

2. Description of the Related Art An image signal is divided into blocks made up of a plurality of pixels, an orthogonal transform is performed in units of the blocks, and a quantization step width corresponding to the position of each transform coefficient,
The technique of quantizing this transform coefficient and performing variable length coding on this quantized data is the basis of international standardization of compression methods for still images and moving images.

In the compression method using the orthogonal transform, the quantization error amount varies depending on how the pixel value in each block varies, and the image quality varies from block to block. Several methods have been proposed in order to reduce such variations in image quality among blocks.

For example, in Japanese Patent Laid-Open No. 3-210878, each block of a decoded image of this coded data and an original image is quantized and coded with a first quantization step width smaller than a target code amount. For each error, the original image is requantized and encoded in order from the block with the largest error with a step size smaller than the first quantization step size, and this requantization and reencoding are performed on the entire image. A method of continuing until the code amount reaches a target code amount is disclosed.

Further, in Japanese Patent Laid-Open No. 3-181229, a parameter is generated by using a transform coefficient value in each block, and a coefficient for multiplying a quantization matrix is changed for a block whose parameter exceeds a threshold value. That method is disclosed.

[0006]

However, in the former method disclosed in Japanese Patent Laid-Open No. 3-210878, after the entire screen is encoded and decoded once, the error between the original image and the original image is small. Must be detected and the blocks must be rearranged in order of increasing error. When such a process is performed, a sorting circuit that rearranges the blocks according to the magnitude of the error, a memory that holds the coordinates of each block after sorting, and a frame memory that holds the decoded image are required.

Further, the quantization step width is reduced again, and the coded bit string when quantized and coded is inserted into the coded bit string that has been quantized and coded using the initial quantization step width. If the coding is variable-length coding, the delimiter of the coded bit string corresponding to each block is stored, or the coded bit string is held for each block and finally 1 It is necessary to have a means of outputting as one serial bit string.

Further, in the latter method disclosed in Japanese Patent Laid-Open No. 3-181229, the actual quantization error is not evaluated as a parameter, but the difference in the image pattern is used as a parameter. , The quantization step width is not always reduced only for the block having a large quantization error. That is, by reducing the quantization step width of an unnecessary block, wasteful information may be increased and the coding efficiency is reduced. Also, if the error from the original image is simply adopted as a parameter to change the quantization step width like the former method, even if this parameter is the same block, it is a distortion that is visually noticeable to human eyes, or it is conspicuous. There is no distortion. That is,
If the quantization step width is simply controlled by the error from the original picture, the coding efficiency will be poor.

The present invention has been made in view of the above points, and generates a parameter capable of detecting a portion in which a visually noticeable distortion is generated without requiring complicated processing, and uses this parameter. It is an object of the present invention to provide an image encoding device and a moving image encoding device capable of improving the visual image quality by changing the quantization step width in block units.

[0010]

In order to achieve the above-mentioned object, the image coding apparatus of the present invention divides an input image signal into blocks composed of a plurality of pixels, and then performs orthogonal transform such as discrete cosine transform, An image coding apparatus that quantizes the orthogonally transformed transform coefficient with a quantization step width corresponding to the position of the transform coefficient in a block, and performs variable-length coding on the quantized data. Error calculation means for calculating an error in each pixel caused by a quantization error when the coefficient is quantized; error selection means for selecting an error for each pixel in the block calculated by the error calculation means in pixel unit; Using the error selected by the error selecting means, a parameter generating means for generating a parameter for judging the change of the quantization step width, and the parameter generating means for generating the parameter. The parameter is characterized by comprising a quantizing step width changing means for changing the quantization step size to the blocks.

Further, the moving picture coding apparatus of the present invention divides the input image signal into macroblocks composed of a plurality of pixels, and then judges whether or not to predict between frames or fields in units of this macroblock, and predicts. If you want to do the above macroblock prediction error, if you do not predict, directly divide the macroblock into several blocks,
A moving image code that performs orthogonal transform such as discrete cosine transform in this block unit, quantizes the orthogonal transform coefficient with a quantization step width corresponding to the position of the transform coefficient in the block, and performs variable length coding on the quantized data. A device, in particular
An error calculating unit that calculates an error in each input pixel with respect to a quantization error when the transform coefficient of the block is quantized, and an error that selects the error for the pixel in the block calculated by the error calculating unit in a pixel unit. Selecting means, parameter generating means for generating a parameter for judging the change of the quantization step width by using the error selected by the error selecting means, parameter generated by the parameter generating means and a set threshold value And a quantization step width changing means for changing the quantization step width in block units.

[0012]

That is, in the image coding apparatus and the moving picture coding apparatus according to the present invention, the error calculating means calculates the error in each input pixel caused by the quantization error when the transform coefficient of the block is quantized, The error selecting means selects the calculated error with respect to the pixel in the block for each pixel,
Parameter change means determines the change of the quantization step width using the selected error, the quantization step width change means, by the generated parameter, for example,
By comparing the parameter and the set threshold, the quantization step width is changed in the block unit.

That is, since the quantization step width is changed, that is, reduced, in the portion where the visually noticeable deterioration is caused by the content of the distortion of the compressed image, the quantization distortion is reduced. Therefore, it is possible to visually generate a decoded image with little deterioration in image quality.

[0014]

Embodiments of the present invention will be described below with reference to the drawings. FIG. 1A is a block configuration diagram of an image coding apparatus according to the first embodiment of the present invention.

First, the input image is input to the discrete cosine transform unit (DCT) 10 and the quantization scale change determination unit 12 as a block image B divided into a block of a plurality of pixels, for example, 8 × 8 pixels. The DCT 10 performs a discrete cosine transform on this input block to obtain a transform coefficient A
Get (i, j). The memory 14 stores this conversion coefficient A (i, j) from the DCT 10. The quantizer 16
The transform coefficient A (i, j) stored in the memory 14 is quantized.

The dequantizer 18 dequantizes the quantized transform coefficient Q (i, j) from the quantizer 16 to generate a transform coefficient A _Q (i, j) containing a quantization error. To do. The subtractor 20 receives the transform coefficient A _Q (i,
j) and the transform coefficient A before quantization stored in the memory 14
The difference ε (i, j) of each transform coefficient is calculated by taking the difference from (i, j).
j) is generated. Inverse Discrete Cosine Transform Unit (IDCT) 2
2 performs inverse discrete cosine transform of the error ε (i, j) from the subtractor 20 to generate the quantized error block image Bε.
The quantization scale change determination unit 12 determines the input block image B
A quantization scale change parameter is generated based on the quantization error block image Bε from the IDCT 22 and is compared with a threshold value to generate a determination signal.

The switch 24 receives the determination signal from the quantization scale change determination unit 12 as a switching control signal, and only when the quantization scale change parameter is smaller than the threshold value, a variable length encoder (VLC). 26 and the quantizer 16 are connected. The VLC 26 has a switch 24.
Variably encodes the quantized value supplied via. The buffer 28 accumulates the encoded data from the VLC 26 and sends it to a communication device or a recording medium (not shown) at a constant rate.
Further, the data occupation amount in the buffer is sent to the quantizer 16.

Next, the operation of the above configuration will be described. First, an input image is a block image B divided into blocks of a plurality of pixels, for example, 8 × 8 pixels, and D
It is input to the CT 10 and the quantization scale change determination unit 12. The DCT 10 performs a discrete cosine transform on this input block to obtain a transform coefficient A (i, j). The transform coefficient A (i, j) is stored in the memory 14 and then supplied to the quantizer 16. In this quantizer 16, FIG.
As shown in, the quantization step width for each transform coefficient is generated by multiplying the quantization matrix weighting each transform coefficient by a quantization scale Q _s that is a scalar,
Each transform coefficient is quantized by this quantization step width. Here, the quantization scale Q _s is determined by the occupancy of the buffer 28 and the determination signal from the quantization scale change determination unit 12.

Transform coefficient Q quantized by the quantizer 16
(I, j) is dequantized by the dequantizer 18,
A transform coefficient A _Q (i, j) including a quantization error is generated. The difference between the transform coefficient A _Q (i, j) and the transform coefficient A (i, j) before quantization is taken by the subtractor 20 to generate an error ε (i, j) of each transform coefficient. .. That is, ε (i, j) = A (i, j) −A _Q (i, j).

This error ε (i, j) is the IDCT
The inverse discrete cosine transform is performed at 22 to generate a quantization error block image Bε. This quantization error block image Bε
And the input block image B are supplied to the quantization scale change determination unit 12 that generates the quantization scale change parameter.

The quantization scale change determination unit 12 is constructed as shown in FIG. 1 (B). That is, each pixel B (i, j) of the input block image B is input, and the selector 121 first determines whether or not the error in this pixel position is adopted in generating the parameter. Specifically, first, a luminance value that is relatively sensitive to a luminance change visually (for example, if the input has a gradation of 8 bits,
(A range of 0 to 180) is selected, and then the pixel value of a plurality of pixels which are at least on the straight line among the adjacent pixels on the upper, lower, left, and right sides of the brightness value is different from the pixel value of the target pixel. Is within a certain set value, that is, if the target pixel is located at a place where the brightness value is an intermediate value of the dynamic range and the brightness change is relatively small, the error of the target pixel is determined. Select for parameter generation.

FIGS. 3A to 3D are conceptual diagrams showing selection criteria for generating this parameter. Each of these figures shows a block, and the shaded area in the figure shows the part where the luminance value is inconsistent with the first selection criterion. In addition, the portions of “⊚” and “x” in the figure respectively indicate pixels, and the pixel of interest now is “⊚”. Here, the difference in luminance value between the peripheral pixel “x” of the target pixel “⊚” and the target pixel “⊚” is obtained, and in the cases of (A) and (C), the target pixel “⊚” is passed through. If the brightness difference is less than or equal to the set threshold value for both of the two pixels “x” that are not on a straight line, and if the cases of (B) and (D), the pixel “x” is 2
The error of the pixel of interest “⊚” is used for parameter generation only when the brightness difference between the pixels is less than or equal to the set threshold.

Generally, there are two characteristic distortions that occur in encoding using the discrete cosine transform. One of them is block distortion in which block boundaries are discontinuous, and the other is noise that occurs around an edge portion of an image or a portion with a large luminance change, which is called mosquito noise. This 2
Both noises are detected due to an unnatural change in the brightness value in the flat part. In order to reduce this noise, it is necessary to detect an error from the original image at the pixel value of the flat part in the brightness value in the range sensitive to the brightness change, and reduce the quantization distortion of the part (block) where this error is large. good. Based on this idea, the quantization scale changing parameter is generated.

A signal for determining whether or not the input pixel B (i, j) is selected by the selector 121 is sent to the switch 122 as a switching control signal. Accordingly, the switch 122 causes the sum of absolute value calculator 123 to calculate the quantization error Bε (i, j) of the input pixel position given from the IDCT 22 only when the input pixel B (i, j) is selected. Switching is controlled so as to input. Then, the absolute value sum calculator 123 calculates the sum of absolute values of the selected quantization error Bε (i, j) for each block unit, and the calculated sum of absolute values is compared with the set threshold value by the comparator 124. , The decision signal of whether it is larger or smaller than the threshold is the quantizer 16 and the switch 2
4 is output.

In the quantizer 16, when the judgment signal indicates that the sum of absolute values is larger than the threshold value, the value Δ (> 0) is calculated from the quantization scale Q _s.
, And the transform coefficient stored in the memory 14 is quantized again by using the generated quantization scale Q _s ′. After that, processings such as inverse quantization, calculation of quantization error of transform coefficient, inverse discrete cosine transform of quantization error, parameter generation and threshold value determination are repeated.

On the other hand, the switch 24 is provided with the quantizer 16 only when the judgment signal from the comparator 124 as the switching control signal indicates that the sum of absolute values is smaller than the threshold value. And VLC 26 are connected. Then, the quantized value at this time is variably encoded by the VLC 26, the encoded data is accumulated in the buffer 28,
It is sent to a communication device (not shown) or a recording medium at a constant rate.

In the buffer 28, the data occupancy in the buffer is sent to the quantizer 16, and the quantizer 16 (see FIG. 4) corresponding to this occupancy can be calculated in the quantizer 16. Is becoming This buffer occupancy is
It is sent to the quantizer 16 in block units or in a unit of several blocks. In other words, the unit for controlling the amount of data is the block unit or the unit of several blocks. The above processing is performed for all blocks of the image signal.

Next, a second embodiment of the present invention will be described.
FIG. 5 is a block diagram of the moving picture coding apparatus according to the second embodiment of the present invention. The moving picture coding apparatus according to the second embodiment has the configuration shown in FIG.
2, subtractor 34, adder 36, prediction mode determiner 38,
An inverse quantizer 40, an IDCT 42, and a motion compensation predictor (MC) 44 are added.

The operation of such a configuration will be described below. First, the input image is divided into macroblocks (for example, 16 × 16 pixels) including a plurality of pixels, and this macroblock MB is input to the switch 30, the subtractor 34, and the quantization scale change determination unit 12. The predictor macroblock PMB subjected to motion compensation prediction from the previous frame decoded image by the MC 44 is further input to the subtractor 34, and the macroblock MB and the predictive macroblock PMB are subtracted to generate a predictive error macroblock εMB. ..
This prediction error macroblock εMB is input to the switch 30, and the prediction output of the prediction mode determiner 38 determines whether the macroblock MB or the prediction error macroblock εMB is D.
It is selectively input to CT10. Prediction mode determiner 38
Then, the variance of pixel values in the macroblock MB, the sum of squares in the prediction error macroblock εMB, and the like are calculated, and the smaller of these values is selected as the macroblock to be coded, which is input to the DCT 10. Switch 3
Switching control of 0 is performed.

The macroblock thus selected is DC
The data is input to T10, and the macroblock is further subdivided into blocks (see FIG. 6) in units (for example, 8 × 8 pixels),
Discrete cosine transform is performed. The transform coefficient A (i, j) obtained by this discrete cosine transform is stored in the memory 14. This transform coefficient is stored in the quantizer 16 in the buffer 2
The quantization step width of each transform coefficient generated by multiplying the quantization scale Q _s determined by the occupied amount of the encoded data in 8 by the weighted quantization matrix of the quantization step width for each transform coefficient. Is quantized by.

The quantized transform coefficient is supplied to the inverse quantizer 18
Transform coefficient A that is dequantized by
_Q (i, j) is generated for macroblocks. This data is subtracted by the transform coefficient before quantization stored in the memory 14 and the subtracter 20 to generate a quantization error Qε (i, j). That is, A (u, v) -AQ (u, v) = _{Q [} epsilon] (u, v) where u and v are indexes indicating horizontal and vertical spatial frequencies. This quantization error is inverse discrete cosine transformed by the IDCT 22, and the error MBQε at each pixel position
Generate (i, j). Here, i and j are indices indicating pixel positions.

The quantizer scale change determining unit 12 receives the data of the macroblock MB and the error MBQε, and generates and determines a parameter for determining whether to change the quantizer scale. The method of generating the parameters of the quantization scale change determination unit 12 is almost the same as that described in the first embodiment described above, and the difference is that the parameters are generated for each block and All that is required is to take the maximum value and use it as a judgment parameter for one macroblock.

When the quantization scale change determination unit 12 determines that the visually noticeable deterioration of the macro block MB currently being processed is large, the determination signal is input to the quantizer 16 and the quantization scale is changed. Is reduced by a predetermined amount (that is, the quantization step width is reduced), and the memory 1
The transform coefficient before quantization stored in 4 is quantized again. Thereafter, the same processing as the above processing is repeated until the quantization scale change determination unit 12 determines that the visually noticeable deterioration in the macroblock MB is small.

Only when it is determined that the quantization scale change determination unit 12 receives the determination signal as the switching control signal for the switch 24 and the quantization scale is not changed, the VLC
26 and the inverse quantizer 40 are connected to the output of the quantizer 16.

When the switch 24 is connected, V
The LC 26 performs variable-length coding on the quantized transform coefficient, and the resulting data is accumulated in the buffer 28 and
The data stored in 8 is output to a recording medium or the like at a constant rate.

Also, the buffer occupancy of the data accumulated in the buffer 28 is sent to the quantizer 16 in units of several macroblocks (this is called a slice) as shown in FIG.
The quantization scale is updated in the quantizer 16. Here, the buffer occupation amount and the quantization scale are shown in FIG.
It is associated as shown in.

Further, the inverse quantizer 40 inversely quantizes the quantized transform coefficient input from the quantizer 16 via the switch 24, and inverse-discrete cosine transforms the result by the IDCT 42 to obtain MB. Generate _Q or εMB _Q.

The prediction mode determiner 38 switches and controls the switch 32 according to the determined mode (whether the macroblock to be encoded is an input signal or a motion compensation prediction error signal). That is, when the determination mode is a mode for selecting a motion compensated prediction error signal, to connect the output adder 36 MC44, adds the IpushironMB _Q predicted macroblock PMB inverse transform which is the output of the MC44 To generate a decoded image macroblock, and when the determination mode is a mode for selecting an input signal, MC44
And the inversely transformed MB _Q are directly stored in the MC 44 as a decoded image macroblock. The accumulated decoded image is used as a reference image for motion compensation prediction of a frame to be processed next. Then, the above processing is repeated.

As described in the first and second embodiments above, the quantization step width is changed according to the content of the distortion of the compressed image, so that a decoded image with little visual deterioration can be generated. become.

[0040]

As described in detail above, according to the present invention,
Generates a parameter that can detect visually noticeable distortions without requiring complicated processing, and changes the quantization step width in block units using this parameter. It is possible to provide an image encoding device that can be improved.

[Brief description of drawings]

FIG. 1A is a block configuration diagram of an image encoding apparatus according to a first embodiment of the present invention, and FIG. 1B is a block configuration diagram of a quantization scale change determination unit in FIG. ..

FIG. 2 is a diagram for explaining a method of generating a quantization step width in the quantizer shown in FIG. 1 (A).

FIGS. 3A to 3D are conceptual diagrams showing selection criteria for generating parameters.

FIG. 4 is a diagram showing a relationship between a data occupation amount in a buffer and a quantization scale.

FIG. 5 is a block configuration diagram of a moving image coding apparatus according to a second embodiment of the present invention.

FIG. 6 shows slices, macroblocks in the input image,
It is a figure which shows the relationship of a block.

[Explanation of symbols]

10 ... Discrete cosine transform unit (DCT), 12 ... Quantization scale change determination unit, 14 ... Memory, 16 ... Quantizer, 1
8, 40 ... Inverse quantizer, 20, 34 ... Subtractor, 22, 4
2 ... Inverse Discrete Cosine Transform Unit (IDCT), 24, 30,
32 ... Switch, 26 ... Variable length encoder (VLC), 2
8 ... Buffer, 36 ... Adder, 38 ... Prediction mode determiner, 44 ... Motion compensation predictor (MC).

Claims (2)

[Claims] 1. An input image signal is divided into blocks each composed of a plurality of pixels, orthogonally transformed, and the orthogonally transformed transform coefficients are quantized with a quantization step width corresponding to the position of the transform coefficient in the block. In an image coding device for variable-length coding quantized data, an error calculating means for calculating an error in each pixel caused by a quantization error when the transform coefficient of the block is quantized, and the error calculating means. Error selecting means for selecting the error for the pixel in the block calculated in step 1 by pixel, and parameter generating means for generating a parameter for judging the change of the quantization step width by using the error selected by the error selecting means. And a quantization step width changing means for changing the quantization step width in the block unit by the parameter generated by the parameter generating means. The image encoding device characterized by the, the equipped. 2. After dividing an input image signal into macroblocks composed of a plurality of pixels, it is determined whether or not to predict between frames or fields in units of this macroblock, and when predicting, the macroblock prediction error is If you do not predict, directly divide the above macroblock into several blocks, perform orthogonal transform in block units,
The orthogonal transform coefficient is quantized with a quantization step width corresponding to the position of the transform coefficient in the block, and the transform coefficient of the block is quantized in a moving picture coding device that performs variable length coding of this quantized data. Error calculation means for calculating the error in each input pixel with respect to the quantization error at the time, error selection means for selecting the error for the pixel in the block calculated by the error calculation means in pixel units, and the error selection means Using the selected error, a parameter generation unit that generates a parameter for determining a change in the quantization step width and a parameter that is generated by the parameter generation unit and a set threshold value are compared to each other. And a quantization step width changing means for changing the quantization step width in units of the blocks.