JPH06284406A - Orthogonal transformation coder - Google Patents

Abstract

PURPOSE:To obtain the orthogonal transformation coder with a high coding efficiency in which a noise component is hardly noticeable. CONSTITUTION:Activity detectors 8, 9 obtain activity for each (picture element) block based on a transformation coefficient from orthogonal transformation means (horizontal DCT 2 and vertical DCT 3) whose dimension differs. A mode discriminator 10 compares detected activities to discriminate the orthogonal transformation method for a coding object. Based on the result of discrimination, a selector 4 is used to select the transformation coefficient to select orthogonal transformation with an optimum dimension to each block and a quantization device 5 and a variable coder 6 execute the coding.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION In recording, transmission, and display devices for processing digital signals, high efficiency coding for efficiently coding and decoding signals with a smaller code amount is used especially in D
The present invention relates to an encoding device (and a decoding device) that performs orthogonal transformation such as CT.

[0002]

2. Description of the Related Art <Orthogonal Transform Coding> A high efficiency coding system using orthogonal transform such as DCT (Discrete Cosine Transform)
It is widely used in standard systems because it can efficiently use image correlation to reduce data. In this case, it is general to perform orthogonal transformation in two dimensions, vertical and horizontal. This is because in one dimension only horizontal or vertical one-way correlation can be used, whereas in two dimensions, both vertical and horizontal can be handled. In addition, a larger transform block is advantageous in that the correlation can be effectively used, but
There is not much difference for x8 pixels or more. On the other hand, since the quantization error spreads over the entire block, it is preferable that the block is visually small, and the processing amount is small. Therefore, it is most common to perform conversion by a block of 8 × 8 pixels of vertical 8th order and horizontal 8th order. Further, in moving image coding, it is general to perform predictive coding between frames in the time direction and spatially use orthogonal transformation for prediction residuals.

<Conventional Encoding Device> FIG. 5 shows the configuration of a conventional encoding device. The image signal supplied from the image input 1 is subjected to 8th-order DCT in the horizontal direction in the horizontal DCT 2 and subsequently to 8th-order DCT in the vertical direction in the vertical DCT 3 for each two-dimensional block of 8 × 8 pixels. The converted signal is
The quantizer 5 quantizes the quantization step width so that an error is not visually noticeable, and the quantization step width is given to the variable-length encoder 6. Most of the coefficients of the quantized signal are 0. In the variable-length encoder 6, the signals in the two-dimensional block state are set to 1 in the order (scan order) as shown in FIG.
It is array-transformed into a dimensional state (so-called zigzag scanning), and the 0 coefficient is its continuous number, the non-zero coefficient is its value,
It is encoded by a variable length code (VLC) such as Huffman code. The output of the variable-length encoder 5 is output as compressed data from the code output 7 to the decoding device.

<Conventional Decoding Device> FIG. 6 shows the configuration of a conventional decoding device corresponding to FIG. The compressed data supplied from the code input 21 is guided to the variable length decoder 22.
The variable length decoder 22 decodes the variable length code into a fixed length code, and the inverse quantizer 23 uses the code as a representative value of quantization. The output of the inverse quantizer 23 is guided to the vertical inverse DCT 24, which inversely transforms the DCT in the vertical direction. Vertical reverse DC
The output of T24 is guided to the horizontal inverse DCT26 and is horizontally D
The inverse transform of CT is performed. As a result, the obtained reproduced image is output from the image output 27.

[0005]

The conventional two-dimensional orthogonal transform coding is efficient at a portion where the intra-image correlation is relatively high, but is not necessarily suitable for a portion where the correlation is low such as an edge portion of the image. The two-dimensional orthogonal transform is not necessarily an appropriate encoding because the tendency becomes particularly strong in the inter-frame prediction residual signal and the intra-image correlation of the residual signal is considerably low.
There is a portion where the efficiency is higher in the one-dimensional conversion only in the horizontal direction or the vertical direction than the two-dimensional conversion. Further, when the two-dimensional conversion is performed, the quantization error spreads in the two-dimensional block, and the noise component around the edge becomes more noticeable. That is, if the coding error is equivalent, one-dimensional conversion with a smaller conversion block is visually desirable. On the other hand, a coding method such as DPCM is visually desirable because noise components are inconspicuous, but the intra-image correlation cannot be used effectively, and the basic efficiency is not sufficient.

In this regard, the applicant of the present invention detects the activity of the input image itself or the finally coded data, and changes the dimension of the orthogonal transform for each block based on the detected activity to determine the code. Has invented and has already applied for an orthogonal transform encoding device with high conversion efficiency and in which the noise component is inconspicuous (Japanese Patent Application No. 4-255870,
"Orthogonal transform encoder and decoder"). Then, the present invention improves the determination and selection means (such as the activity detection method) of the optimum orthogonal transformation method to detect the activity with a simpler structure with high accuracy, improve the coding efficiency, and make the noise component noticeable. It's difficult.

[0007]

In order to achieve the above object, for example, as shown in FIG. 1, there is provided an encoding device for performing orthogonal transformation on a plurality of dimensions of an image signal, and the dimensions for orthogonal transformation. A plurality of orthogonal transforming means (horizontal DCT2, vertical DCT3) having different combinations, and a plurality of activity detecting means (activity detectors 8 and 9) for obtaining an activity based on transform coefficients respectively output from the plurality of orthogonal transforming means. Determination means (mode determiner 10) for comparing the activities obtained by the plurality of activity detection means and determining the orthogonal transformation method to be encoded, and determination information (mode information) output from the determination means And a means (selector 4) for selecting a transform coefficient based on the above. .

Further, as shown in FIG. 2, for example, a coding device for performing orthogonal transform on a plurality of dimensions of an image signal, wherein a plurality of orthogonal transform means (horizontal DCT 2, The vertical DCT 3) and the transform coefficients respectively output from the plurality of orthogonal transform means,
A plurality of quantizing means (quantizers 5, 31) for quantizing,
A plurality of activity detecting means (activity detectors 8 and 9) for obtaining an activity based on the quantized transform coefficient output from each of the plurality of quantizing means.
And a determination unit (mode determination unit 10) that determines the orthogonal transform method to be encoded by comparing the activities obtained from the plurality of activity detection units, and the determination information (mode information) output from the determination unit. ),
The present invention provides an orthogonal transform coding device characterized by comprising means (selector 4) for selecting a quantized transform coefficient.

[0009]

According to the above-mentioned orthogonal transform coding device, the activity is detected by the transform coefficients from a plurality of orthogonal transform means or by the quantized transform coefficient, and the orthogonal transform method is set as the coding target. (Orthogonal transformation method with different combinations of dimensions) is selected and selected.

[0010]

DESCRIPTION OF THE PREFERRED EMBODIMENTS <Embodiment (Encoding Device 1)> FIG. 1 is a block diagram showing a first embodiment of an orthogonal transform encoding device of the present invention. The same parts as those in the conventional example shown in FIG. The encoding device of FIG. 5 is the activity detector 8,
9 is different in that a mode determiner 10, a selector 4, and a mode output 11 are added. In addition, the above-mentioned prior application “Orthogonal transform encoding device and decoding device” (Japanese Patent Application No. 4-2
55870) is different only in the mode determination method of the orthogonal transformation method.

In FIG. 1, the image signal supplied from the image input 1 is a horizontal DCT in a block state of 8 × 8 pixels.
DCT is performed in the horizontal direction by 2 and applied to the vertical DCT 3, the activity detector 8 and the selector 4. Vertical DC
At T3, the signal DCTed in the horizontal direction is subjected to DCT in the vertical direction, and the signal subjected to the two-dimensional DCT is given to the activity detector 9 and the selector 4. Here, in the case of conversion only in the horizontal direction, the DCT block has a one-dimensional 8
Although it becomes pixels, it is convenient to bundle eight DCT blocks in the vertical direction and handle them in a block state of 8 × 8 pixels for the convenience of switching to two-dimensional processing. This state is shown in FIGS. 7 (A) and 7 (B).

The activity detector 8 obtains the activity of the signal DCTed only in the horizontal direction. The activity is, for example, the sum of absolute values of the coefficients, and is expressed by the formula (1).
Is indicated by A1d. C1d _xy is x-th coefficient value of the vertical y-th horizontal DCT in equation (1), C1d _0y is a DC component.
Although there are eight DC components in a block of 8 × 8 pixels, a simple sum of absolute values is inappropriate as an activity, and therefore a difference is calculated between DC values. In addition, the first one that does not make a difference is excluded from the addition.

The activity detector 9 obtains the activity of the horizontal and vertical two-dimensional DCT signals. The activity is similarly obtained by A2d shown in the equation (2). In equation (2), C2d _xy is the horizontal x, vertical y-th coefficient value of the two-dimensional DCT, and C2d ₀₀ is the DC component. Further, one DC component as shown in Expression (2) is excluded.

[0014]

[Equation 1]

[0015]

[Equation 2]

When the signal to be coded is not the image signal itself but the residual signal of inter-picture prediction, the DC component can be regarded as substantially the AC component, so that the absolute values of all the coefficients are simply added. Just do it. Specifically, the activity detector 8 uses equation (3) instead of equation (1), and the activity detector 9 uses equation (2) instead of
Formula (4) is used.

[0017]

[Equation 3]

[0018]

[Equation 4]

Here, the unit for switching the processing is DCT.
The block does not necessarily have to be the same as the block (8 × 8 pixels), and a plurality of DCT blocks may be bundled. Especially, when sub-sampling of color difference signals is performed on a color image, a plurality of luminance signal blocks and one color difference signal block form a pair, which may be more convenient in some cases. The horizontal DCT activity obtained by the activity detector 8 and the two-dimensional DCT activity obtained by the activity detector 9 are provided to the mode determiner 10 for each block.

The mode determiner 10 compares both activity values and determines which one is smaller, and obtains mode information. In comparison, add about 1 per coefficient to either activity value, or 1.2
The selection ratio may be adjusted by giving a slight offset such as multiplying by about 5. Mode information is mode output 11
Is output to the decoding device and is also given to the selector 4, the quantizer 5 and the variable length encoder 6. The selector 4 selects a transform coefficient according to the mode information and supplies it to the quantizer 5.

In the quantizer 5, if uniform quantization is performed on each coefficient, the processing is single, but if weighting corresponding to visual characteristics is performed on each DCT coefficient, conversion is performed. Processing differs depending on the method. The weighting of quantization has a two-dimensional characteristic when it is two-dimensional according to the mode information, but has a one-dimensional characteristic that changes only in the horizontal direction when it is only horizontal. The output of the quantizer 5 is the variable length encoder 6
Is encoded with. Array conversion (scan) with variable length coding
Is different in the two-dimensional case and in the horizontal case only. The two-dimensional mode information is the same as the conventional example as shown in FIG. 8A, but the horizontal case is as shown in FIG. 8B. In this way, by processing the one-dimensional conversion by bundling eight,
The efficiency of 0 run length coding can be improved, and the same VLC
You can use the table. The code compressed by the variable length encoder 6 is output from the code output 7 to the decoding device.

In this way, the activity detectors 8, 9
Then, the activity of the signal DCTed only in the horizontal direction and the activity of the signal horizontal and vertical two-dimensional DCT are obtained, that is, from the plurality of orthogonal transform means (horizontal DCT2, vertical DCT3). Since the activity is detected by the transform coefficient and the orthogonal transform method to be encoded is determined and selected, the configuration is simpler than the case where the activity is detected from the input image itself as in the previous application. Therefore, high-speed processing is possible, and the accuracy is higher than that in the case of detecting the activity from the finally encoded data. This is a horizontal DC
While the transform coefficients from T2 and vertical DCT3 directly indicate the activity required for the determination and selection of the orthogonal transform method, the input image itself requires a special means for activity detection, and The reason is that the finally encoded data is improperly weighted by encoding (variable length encoding). In the present embodiment, only the two-dimensional and horizontal ones are shown, but three types of two-dimensional and vertical only, vertical only and horizontal only, two-dimensional, vertical only and horizontal only, and even when orthogonal transformation is not performed Including adaptive processing is also possible. Further, as described above, it may be applied to the residual signal of inter-picture prediction.

<Embodiment (Encoding Device 2)> FIG. 2 is a block diagram showing a second embodiment of the orthogonal transform encoding device of the present invention. The same parts as those in the first embodiment of FIG. 1 are designated by the same reference numerals. The encoder differs from the encoder of FIG. 1 in that a quantizer 31 is added and the activity detectors 8 and 9 are provided after the quantizers 5 and 31, respectively.

In FIG. 2, the image signal given from the image input 1 is DCTed in the horizontal direction by the horizontal DCT 2 and given to the vertical DCT 3 and the quantizer 5. Vertical DC
At T3, the signal DCTed in the horizontal direction is subjected to DCT in the vertical direction, and the signal subjected to the two-dimensional DCT is given to the quantizer 31. The quantizer 5 weights each DCT coefficient with a one-dimensional horizontal-only characteristic corresponding to visual characteristics during quantization, and the quantized coefficient is applied to the activity detector 8 and the selector 4. Given. The quantizer 31 performs weighting corresponding to the visual characteristic with a two-dimensional characteristic, and the quantized coefficient is given to the activity detector 9 and the selector 4.

The operations of the activity detector 8 and the activity detector 9 are the same as those in FIG. 1, but since the processing target is weighted and quantized, the weight becomes lighter as the frequency component becomes higher. Since the weighting of the quantization is taken into consideration in this way, the judgment is more suitable for the actual encoding. Horizontal DC required by activity detector 8
The activity of T and the activity of the two-dimensional DCT obtained by the activity detector 9 are given to the mode determiner 10 for each block.

The mode determiner 10 compares both activity values and determines which one is smaller, and outputs mode information for orthogonal transformation. The mode information is output from the mode output 11 to the decoding device and is also given to the selector 4 and the variable length encoder 6. The selector 4 outputs the output of the quantizer 5 and the quantizer 3 according to the mode information.
One of the outputs of 1 is selected and supplied to the variable length encoder 6. The variable-length encoder 6 encodes by changing the scan according to the mode information. The code compressed by the variable length encoder 6 is output from the code output 7 to the decoding device.

<Embodiment (Encoding Device 3)> FIG. 3 is a block diagram showing a third embodiment of the orthogonal transform encoding device of the present invention. The same parts as those in the first embodiment of FIG. 1 are designated by the same reference numerals. The configuration differs from the encoding apparatus of FIG. 1 in that the coefficient weighting units 41 and 42 are provided. Although the basic operation of the embodiment of FIG. 3 is the same as that of the embodiment of FIG. 1, the coefficient is weighted when the activity is obtained.

In FIG. 3, a horizontal DCT 2 and a vertical DCT
The operation of 3 is the same as that of FIG. The output of the horizontal DCT 2 is given to the coefficient weighting unit 41 and the vertical DCT 3, and the output of the vertical DCT 3 is given to the coefficient weighting unit 42. The coefficient weighting unit 41 divides the coefficient value by a value according to the weighting at the time of quantization of the two-dimensional DCT. Similarly, the coefficient weighting unit 42 divides the coefficient value by a value according to the weighting at the time of quantization of the horizontal DCT.
That is, the processing is the same as that performed by the quantizer 5 with a sufficiently fine quantization step width. As for the divided coefficient, the output of the coefficient weighting device 41 is sent to the activity detector 9,
The output of the coefficient weighting device 42 is provided to the activity detector 8.

As shown in FIG. 2, when the judgment is made by the quantized coefficient,
Whereas fine coefficient values are not quantized, in FIG. 3 weighted coefficients can be used while they are stored. The operations of the activity detectors 8 and 9, the mode determiner 10, the selector 4, the quantizer 5, and the variable length encoder 6 are the same as those in FIG. 1, and the compressed code is output from the code output 7 to the decoding device. .

<Embodiment (Decoding Device)> FIG. 4 shows the structure of a decoding device corresponding to FIGS. 1, 2 and 3. The same parts as those in the conventional example of FIG. 6 are designated by the same reference numerals. 6 is different from the decoding apparatus of FIG. 6 in that a mode input 28 and a selector 25 are provided and the operations of the variable length decoder 22 and the inverse quantizer 23 are different. The compressed data supplied from the code input 21 is guided to the variable length decoder 22. On the other hand, the mode information input from the mode input 28 is the variable length decoder 22 and the inverse quantizer 2
3, provided to the selector 25.

The operations of the variable length decoder 22 and the inverse quantizer 23 are basically the same as those of the conventional example, but the scan is changed in the variable length decoder 22 and the inverse quantizer 23 is changed according to the mode information. The weighting for the coefficients is changed. The output of the inverse quantizer 23 is the vertical inverse DCT 24 and the selector 25.
The vertical inverse DCT 24 inversely transforms the DCT in the vertical direction. The selector 25 is controlled by the mode information, and in the block of the two-dimensional DCT, the vertical inverse DCT2
4 output is selected, and in the case of horizontal only, the inverse quantizer 2
The output of 3 is selected and led to the horizontal inverse DCT. In the horizontal inverse DCT, inverse conversion of DCT is performed in the horizontal direction. As a result, in the block of the two-dimensional DCT, both the vertical and horizontal inverse DCT are performed, and in the case of only the horizontal, the horizontal inverse DCT is performed.
Only is performed, and the obtained reproduced image is output from the image output 27.

[0032]

According to the orthogonal transform coding device of the present invention, the dimension of the orthogonal transform is changed for each block and the optimum transform method is selected, so that it can be compared with a device using a single transform method. Optimum encoding can be performed locally and the encoding efficiency of the entire image can be improved. Furthermore, since one-dimensional conversion is used at edges and the like, quantization noise becomes less noticeable, and subjective image quality becomes better than error reduction.

Further, since the determination of the transform method is performed by the activity of the transform coefficient subjected to the orthogonal transform or the transform coefficient subjected to the orthogonal transform / quantization, the activity can be easily detected and its accuracy is also high. As described above, the orthogonal transform coding device of the present invention has extremely excellent practical effects.

[0034]

[Brief description of drawings]

FIG. 1 is a block diagram showing a first embodiment of an orthogonal transform coding device according to the present invention.

FIG. 2 is a block diagram showing a second embodiment of the orthogonal transform encoding device according to the present invention.

FIG. 3 is a block diagram showing a third embodiment of an orthogonal transform coding device according to the present invention.

FIG. 4 is a block diagram showing an orthogonal transform decoding device corresponding to the orthogonal transform coding device according to the present invention.

FIG. 5 is a block diagram showing a conventional example of an orthogonal transform encoding device.

FIG. 6 is a block diagram showing a conventional example of an orthogonal transform decoding device.

FIG. 7 is a diagram showing a configuration of a two-dimensional and a horizontal one-dimensional DCT block.

FIG. 8 is a diagram showing the order of two-dimensional and horizontal one-dimensional array conversion (scan order).

[Explanation of symbols]

1 ... Image input, 2 ... Horizontal DCT, 3 ... Vertical DCT, 4 ...
Selector, 5, 31 ... Quantizer, 6 ... Variable encoder, 7
... code output, 8, 9 ... activity detector, 10 ... mode determiner, 11 ... mode output, 21 ... code input, 22
... variable length decoder, 23 ... inverse quantizer, 24 ... vertical inverse D
CT, 25 ... Selector, 26 ... Horizontal inverse DCT, 27 ... Image output, 28 ... Mode input, 41, 42 ... Coefficient weighting device.

Claims (5)

[Claims] 1. An encoding device for orthogonally transforming a plurality of dimensions of an image signal, wherein a plurality of orthogonal transforming means having different combinations of dimensions for orthogonal transforming are output from the plurality of orthogonal transforming means, respectively. A plurality of activity detecting means for obtaining an activity based on the transform coefficient, a judging means for comparing the activities obtained by the plurality of activity detecting means, and judging an orthogonal transform method to be encoded, and an output from the judging means. An orthogonal transform coding device, comprising: means for selecting a transform coefficient based on the determined determination information. 2. An encoding device for orthogonally transforming a plurality of dimensions of an image signal, the plurality of orthogonal transforming means having different combinations of dimensions for orthogonal transforming, and the plurality of orthogonal transforming means respectively outputting. A plurality of quantizing means for quantizing the transform coefficient, a plurality of activity detecting means for obtaining an activity based on the quantized transform coefficients respectively output from the plurality of quantizing means, and a plurality of activity detecting means It comprises a determination means for comparing the activities obtained from the above and determining an orthogonal transformation method to be encoded, and a means for selecting a quantized transform coefficient based on the determination information output from the determination means. An orthogonal transform coding device characterized by the above. 3. The activity detecting means is adapted to calculate the activity by using the difference between the DC components when a plurality of DC components of the orthogonal transformation are present in the block for which the activity is determined. Alternatively, the orthogonal transform encoding device according to claim 2. 4. The orthogonal transform coding apparatus according to claim 1, wherein the activity detecting means weights the transform coefficient according to the frequency to obtain the activity. 5. The encoding device according to claim 1 or 2, further comprising encoding means for performing encoding corresponding to the orthogonal transform method selected and decided based on the decision information output from the decision means. Orthogonal transform coding device.