JPH06188742A - Conversion encoding and decoding device - Google Patents

Abstract

PURPOSE:To obtain a high-efficiency encoding device and a decoding device for a signal. CONSTITUTION:A block deciding circuit 2 sections an input signal into blocks of various size according to its local characteristics and an unequal division converting and encoding circuit 3 performs band unequal division type conversion and encoding for the signal. An effective coefficient deciding circuit 4 classifies the conversion coefficients outputted from the unequal division converting and encoding circuit 3 into effective coefficients and ineffective coefficient based on the blocks sectioned by the block deciding circuit 2 and outputs only the effective coefficients.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a transform coding device and a decoding device.

[0002]

2. Description of the Related Art Here, a discrete cosine transform (hereinafter referred to as "DC
T ”) is used in a transform coding apparatus for image signals. Figure 9 DIGI of General Instrument Corporation
FIG. 3 is a block circuit diagram of an image signal transform coding apparatus combining motion compensation prediction and DCT described in a document of CIPHERTM HDTV SYSTEM, 8 June 1990. In the figure, 901
Is an input terminal for the color difference signal U, 902 is an input terminal for the color difference signal V, and 903 is an input terminal for the luminance signal Y. A thinning circuit 904 is connected to the input terminal 901, and a thinning circuit 905 is connected to the input terminal 902. A multiplexer 906 is connected to the thinning circuit 904, the thinning circuit 905, and the input terminal 903, and is connected to one input of the subtractor 907. Reference numeral 908 denotes a DCT arithmetic circuit connected to the subtractor 907, 909 denotes a quantizer connected to the DCT arithmetic circuit 908, 910 denotes an inverse quantizer connected to the quantizer 909, and 911 denotes an inverse quantizer 910. The connected inverse DCT arithmetic circuit is connected to one input of the adder 912. Reference numeral 913 is a frame memory connected to the adder 912, 914 is a motion vector detection circuit connected to the multiplexer 906 and the frame memory 913, and 915 is a motion compensation circuit connected to the frame memory 913 and the motion vector detection circuit 914. , Subtractor 907 and adder 912 are connected to the other inputs. Reference numeral 916 is a variable length coding circuit connected to the quantizer 909, and 917 is an output terminal connected to the variable length coding circuit 916.

Next, the operation will be described. The DCT arithmetic circuit 908 converts the prediction error signal output from the motion compensation predictive coding circuit, which has been executed through the motion vector detection circuit 914, the motion compensation circuit 915, and the subtractor 907, into blocks of 8 horizontal pixels and 8 vertical lines. Division is performed and DCT is performed for each block. The DCT transforms pixels in a block into a kind of frequency domain (hereinafter referred to as "transform domain"), and divides the horizontal and vertical directions from low frequency to high frequency into eight stages. The DCT coefficient is the amount of the signal allocated in each of the sub-regions divided in the conversion region divided into a total of 64 horizontal and vertical 8 regions. The DCT coefficient is quantized by the quantizer 909. The DCT coefficients are finely quantized both in the horizontal and vertical regions in the low frequency conversion region and coarsely quantized in the high frequency conversion region. Since the image signal has such a property that power is concentrated in the low frequency conversion region and the power amount is small in the high frequency conversion region, the high frequency component of the quantized DCT coefficient is 0. There are many things that become. The DCT coefficient in which the number of 0 is increased is subjected to variable length coding with respect to 64 coefficients per block, that is, all coefficients by the variable length coding circuit 916, and output. As a variable-length code, the number of consecutive 0s as the quantization value (hereinafter referred to as "zero run length")
And a non-zero quantization value immediately after that,
Determine the length of the codeword according to the probability of this combination 2
It uses what is called a three-dimensional Huffman code.

On the decoding side, the variable-length codeword is decoded, inverse quantization and inverse DCT are performed, and the prediction error signal is decoded. The prediction error signal restores the original image signal by motion compensation predictive decoding.

[0005]

In the transform coding, as described in the conventional example, there are many cases where the value after quantization is 0 in transform coefficients of particularly high frequency. Therefore, conventionally, all the transform coefficients have been output by a two-dimensional Huffman code having a code word that is a combination of a zero run length and a nonzero quantized value immediately after that.

However, if the number of 0s to be encoded can be reduced with respect to the value of the quantized transform coefficient being 0, that is, if the number of 0s transmitted from the encoder can be reduced, It is possible to perform highly efficient encoding with a low information transmission rate, a high compression rate. This is because if the coefficient value can be predicted based on the local property of the signal in the spatial domain (and / or the temporal domain), which is particularly 0 in the transform domain, in the spatial domain. By encoding and transmitting the local property of the above, it is possible to realize it by not transmitting the one whose coefficient value can be predicted to be 0 based on the property of this spatial region.

However, conventional DCT and Hadamard transform,
In transform coding such as band equalization sub-band coding, in a block obtained by dividing the original signal for conversion, only a portion where the signal is present has a sharp change in signal level and a high frequency component exists, For other parts, even if there are local properties of the signal such as flat and low frequency components only, predict which conversion coefficient value will be 0 based on this local property. Was difficult. This is because these transform codings have the same length of transform bases (or may be called coding filters) by which low-frequency transform components and high-frequency transform components are multiplied. Exists, and in the other parts only low-frequency components exist,
This is because a high frequency component that exists only in a part is multiplied by all the conversion bases (or the encoding filters), and thus it is not possible to deny the possibility that any conversion coefficient takes a non-zero value.

The present invention has been made in order to solve the above-mentioned problems, and it is possible to predict the value of the transform coefficient which becomes 0 based on the local property in the spatial domain. An object of the present invention is to obtain a transform coding device / decoding device having a higher compression rate than conventional by not transmitting the transform coefficient predicted to be 0 that is possible from the coding device.

[0009]

According to a first aspect of the present invention, a transform coding apparatus divides a signal into blocks of various sizes and outputs transform coefficients corresponding to the positions of the blocks. It is adaptively controlled according to the size of the block.

A transform coding apparatus according to a second aspect of the invention is
Motion compensation predictive coding with variable block size is performed on the image signal, and the output of the transform coefficient corresponding to the position of each block used for this motion compensation predictive coding is
It is adaptively controlled according to the size of this block.

A transform coding apparatus according to a third aspect of the invention is
The motion vector is encoded based on the block structure.

The transform coding apparatus according to the invention of claim 4 is
The block structure information is subjected to variable length coding such as entropy coding including Huffman coding and arithmetic coding.

A transform coding apparatus according to a fifth aspect of the present invention is
The quantization step width of each transform coefficient is adaptively changed according to the block structure.

According to a sixth aspect of the present invention, there is provided a conversion decoding device.
This is to detect which invalid coefficient is based on the block structure information, substitute a constant in this invalid coefficient, and perform transformation decoding together with the valid coefficient.

According to a seventh aspect of the present invention, there is provided a conversion / decoding device comprising:
Detecting which is an invalid coefficient or an effective coefficient having a certain constant value, and this invalid coefficient or an effective coefficient having a certain constant value is not present at a specific position with respect to the decoding filter. The decoding is performed by multiplying a decoding filter by using an invalid coefficient as a constant and other effective coefficients, and an invalid coefficient or an effective coefficient having a predetermined constant value is set at a specific position with respect to the decoding filter. , The decoding value by multiplication with the decoding filter is always a constant.

According to an eighth aspect of the present invention, there is provided a conversion / decoding device comprising:
The block structure information is obtained based on the received motion vector.

[0017]

The effective coefficient determining means in the invention of claim 1 is
By controlling the output transform coefficient according to the block size, only the transform coefficient that is suitable for the decoding of the signal and that is adapted to the local property of the signal is transmitted.

The non-uniform division transform coding means in the second aspect of the present invention is combined with the motion compensation predictive coding means of variable block size, so that the block structure information is used for both motion compensated prediction and transform coding. .

The motion vector encoding means in the third aspect of the present invention encodes the motion vector based on the block structure, so that the encoded motion vector also includes the block structure information. There is no need to separately transmit the structural information.

According to the fourth aspect of the invention, the block information coding means reduces the code amount of the block structure information to near the entropy by performing variable length coding on the block structure information.

The quantizing means in the invention of claim 5 indicates by what quantizing step width each transform coefficient is quantized by adaptively changing the quantizing step width according to the block structure. Information need not be newly transmitted.

The invalid coefficient substituting means in the sixth aspect of the present invention makes it possible to perform transform decoding by substituting a constant in the invalid coefficient.

In the non-uniform division transform decoding means in the invention of claim 7, when the invalid coefficient or the effective coefficient having a predetermined value is located at a specific place with respect to the decoding filter, By always setting the value of the multiplication of the invalid coefficient and the decoding filter as a certain constant, the influence of the transform coefficient existing around the invalid coefficient and the like is removed to perform the decoding.

According to the block information decoding means in the invention of claim 8, since the motion vector code word output from the motion vector coding means of claim 3 also includes the block structure information, Extract block structure information from the motion vector code word.

[0025]

EXAMPLES Example 1. Embodiment 1 of the present invention will be described below with reference to the drawings. FIG. 1 is a block circuit diagram of the transform coding apparatus according to the first embodiment and corresponds to the DCT operation circuit 908 in FIG. In the figure, 1 is a signal input terminal, 2
Is a block determination circuit connected to the input terminal 1, 3 is a non-uniform division conversion coding circuit connected to the input terminal 1, 4 is an effective coefficient connected to the block determination circuit 2 and the non-uniform division conversion coding circuit 3. The determination circuit 5 is an output terminal connected to the block determination circuit 2, and 6 is an output terminal connected to the effective coefficient determination circuit 4.

Next, the operation will be described. The block determination circuit 2 divides a flat signal portion, which does not change much, into large blocks, and a portion having a gentle signal level change, into a medium-sized block, and a portion having a sharp signal level change. Then, it is divided into blocks of several sizes according to the local characteristics of the signal, like division into small blocks. Information on the separated blocks is output from the output terminal 5 as block structure information. The division of this block is performed in order to see a rough local property of the signal, and is not intended to divide the signal into blocks and perform closed processing in each block. In fact, in transform coding such as wavelet transform, transform bases are long, and adjacent transform bases partially overlap each other, so that blocks cannot be completely divided.

The band non-uniform division type transform coding will now be described. Typical examples of the band non-uniform division type transform coding include wavelet transform and band non-uniform division type sub-band coding. Transform coding is performed by dividing a signal into N frequency bands such as a low frequency band, a medium frequency band and a high frequency band using a low pass filter, a band pass filter, a high pass filter, etc. The sampling points of the signals in the respective regions are thinned according to the narrowed bandwidths.

Further, in order for this transform coding to be a non-uniform band division type, the frequency band width passed by the low-pass filter is narrowed, and the pass frequency band width is widened as the frequency band passed is increased. There is a need to. This is shown in FIG. The filter has a larger number of taps as it passes lower frequencies, and has a smaller number of taps as it passes higher frequencies. In accordance with this, the signal in the low frequency region has a narrow bandwidth, so the sampling rate is large, that is, the number of sampling points is small, and the bandwidth becomes wider in the high frequency region. Reduce the rate. This is shown in FIG.
It shows in (b).

To summarize the above, the non-uniform band-type transform coding means that the frequency domain resolution is higher, the spatial (or time) domain resolution is lower, and the higher the lower frequency components of the signal are. It can be said that this is a transform coding in which the frequency domain resolution is lowered and the spatial (or time) domain resolution is increased as the number of domain components increases.

With respect to the signal transform-coded by the non-uniform division transform-encoding circuit 3, each transform coefficient is divided into a valid coefficient and an invalid coefficient by the valid coefficient judging circuit 4. Here, in the effective coefficient determination circuit 4, the block determination circuit 2 determines that a large block portion is a flat signal portion that does not change so much. Therefore, among the conversion coefficients corresponding to the block positions,
Since the values of the mid-range transform coefficient and the high-range transform coefficient are almost 0, these transform coefficient values are not output, only the low-range transform coefficient is output and transmitted.
Since there is a gradual change in the signal level, among the conversion coefficients corresponding to the position of the block, the values of the high-frequency conversion coefficients will be almost 0, and these conversion coefficient values will not be output. The mid-range transform coefficient is output and transmitted, and the part where the block is small is the part where the signal level changes abruptly. Therefore, of the transform coefficients that correspond to the position of the block,
In order to output and transmit all the transform coefficients, it is determined how many high-frequency transform coefficients to output according to the block size by classifying into several types.

Here, the transform coefficient corresponding to the position of the block means that the center of gravity of the transform base multiplied by the signal for obtaining the transform coefficient is in or near the block. Also, in band non-uniform division type transform coding such as wavelet transform, the lower the low band, the longer the length of the transform base and the greater the distance between corresponding positions of adjacent transform coefficients. That is, the lower the low-frequency transform coefficient, the smaller the number of sample points, and the signals are sparsely arranged. Therefore, when medium or small blocks are adjacent to each other, the low-frequency transform coefficients corresponding to the respective blocks are often the same. In such a case, it is not necessary to output the low-frequency transform coefficient only once and output the same coefficient twice or more.

The method of dividing a signal into blocks by using the local property of the signal and not outputting the less important ones of the transform coefficients is a transform coding in which a higher band transform than a lower band transform base such as a wavelet transform is performed. This can be realized by using band non-uniform division type transform coding with a short base length. For a signal that has a sharp local signal level change, the high-frequency conversion component has a large power value only in the part where the level changes, and the high-frequency conversion base is short in other flat parts. , The conversion base is not applied to the part where the level changes, and the power value of the high frequency conversion coefficient becomes small. Therefore, in the flat portion, the restored signal does not deteriorate even if the value of the high-frequency transform coefficient is assumed to be 0 and output is not performed.

In this way, the effective coefficient is output through the effective coefficient judgment circuit 4 and the output terminal 6, and the block structure information used for the judgment of the effective conversion coefficient is output from the output terminal 5.

Example 2. Embodiment 2 of the present invention will be described below with reference to the drawings. FIG. 2 is a block circuit diagram of the transform coding apparatus according to the second embodiment. In the figure, the non-uniform division conversion coding circuit 3, the effective coefficient determination circuit 4, and the output terminal 5 are the same as those in the first embodiment shown in FIG. Reference numeral 7 is an image signal input terminal, 8 is a block size variable motion compensation prediction coding circuit connected to the input terminal 7, 9 is an output terminal connected to the effective coefficient determination circuit 4, and 10 is a block size variable motion compensation prediction code. It is an output terminal connected to the digitization circuit 8.

Next, the operation will be described. The coding apparatus according to the second embodiment combines the coding apparatus according to the first embodiment with a variable block size motion compensation prediction coding circuit 8 and uses block structure information as motion compensation prediction with variable block size and valid / invalid transform coefficients. It is characterized by using it for both judgments. Only for motion compensated prediction with variable block size,
Alternatively, if the block structure information is used only for transform coding, the motion compensation prediction error signal power can be reduced, or the transform coefficient to be output can be reduced, but in any case, the block structure information itself is transmitted. Because you need
The efficiency is reduced by the transmission of this information. By using the block structure information for both motion-compensated prediction and determination of valid / invalid of the transform coefficient, it is possible to suppress deterioration in efficiency due to transmission of the block structure information.

In the motion compensation prediction with variable block size, for example, motion compensation can be accurately performed in a large block, and when the prediction error signal power is small, motion compensation prediction is performed in a large block and the motion is performed in units of large blocks. If the accuracy is not sufficient in compensation and the prediction error signal power becomes large, motion compensation prediction should be performed by obtaining the motion vector of each small block in small block units only for that part. In addition, the block size for motion compensation prediction is changed to several types according to the magnitude of the motion compensation prediction error signal power.

When such motion-compensated prediction is performed, the motion-compensated prediction error signal in a large block becomes a flat signal with low power, and as the block becomes smaller, prediction is performed rather than performing motion-compensated prediction in a large block. Although the error power is small, the prediction error power is still small, and the prediction error signal has a signal level change.

As described above, the variable block size motion compensation predictive coding circuit 8 performs motion compensation predictive coding on the image signal, outputs the motion vector at the output terminal 10, and the prediction error signal at the non-uniform division conversion coding circuit 3. To the effective coefficient determination circuit 4 and the output terminal 5. The information of the block structure used for the motion compensation prediction is output to the effective coefficient determination circuit 4 and the output terminal 5.

The non-uniform division transform coding circuit 3 performs band non-uniform division transform coding such as wavelet transform on the prediction error signal output from the block size variable motion compensation predictive coding circuit 8. The conversion coefficient is output. In the effective coefficient determination circuit 4, the conversion coefficient output from the non-uniform division conversion encoding circuit 3 is converted into an effective coefficient and an invalid coefficient based on the block structure information output from the block size variable motion compensation prediction encoding circuit 8. And output only the effective coefficient. The block structure information used for determining the effective coefficient has the same property as the block structure of the first embodiment due to the property of the prediction error signal output by the motion compensation prediction with variable block size. That is, the larger the block is, the flatter the prediction error signal is in that part, and there is less power in the high-frequency transform component. Conversely, the smaller the block is, the larger the prediction error signal is in that part, and the lower-frequency transform component The power is widely distributed from to the high frequency conversion component. By utilizing this property, it is possible to determine the effective coefficient and the invalid coefficient as in the first embodiment. That is, the larger the block is, the lower the transform coefficient of the transform coefficients corresponding to the position of the block is output, and the smaller the block is, the more corresponding to the position of the block. Among the conversion coefficients, the higher conversion coefficient is output.

In this way, the effective coefficient is outputted from the effective coefficient judging circuit 4 and the output terminal 9, and the block structure information used for both the block variable motion compensation prediction and the judgment of the effective transform coefficient is outputted from the output terminal 5. Is output.

Example 3. Embodiment 3 of the present invention will be described below with reference to the drawings. FIG. 3 is a block circuit diagram of the transform coding apparatus according to the third embodiment. In the figure, the input terminal 7, the block size variable motion compensation predictive coding circuit 8, the unequal division conversion coding circuit 3, the effective coefficient determination circuit 4, and the output terminal 9 are the same as those in the second embodiment shown in FIG. is there. Reference numeral 11 is a motion vector coding circuit connected to the block size variable motion compensation predictive coding device 8, and 12 is a motion vector coding circuit 1.
It is an output terminal connected to 1.

Next, the operation will be described. The coding apparatus of the third embodiment is obtained by adding the motion vector coding circuit 11 to the coding apparatus of the second embodiment. The motion vector coding circuit 11 can include the information of the block structure in the motion vector code word in order to code the motion vector according to the block structure, and separately code the block structure information. Therefore, it is possible to prevent the deterioration of efficiency due to the transmission of the block structure information.

An example of a motion vector coding method according to the block structure is shown below. First, the motion vector in the largest block when performing motion compensation prediction is encoded. Next, the motion vector in the second largest block is encoded. A block having this size is assigned a codeword for a motion vector for which a new motion vector has not been obtained, and this new motion vector is coded for a block for which a new motion vector has been obtained.

As an example of encoding a new motion vector, the new motion vector and the one previously encoded,
There is also a method of encoding a difference with a motion vector of a block that is one step larger. The above operation is continued for a new motion vector allocation, and nothing is coded for a new motion vector allocation. Such an operation is performed until there is no motion vector that needs to be encoded.

In this way, the motion vector including the block structure information is output from the output terminal 12.

Example 4. Embodiment 4 of the present invention will be described below with reference to the drawings. FIG. 4 is a block circuit diagram of the transform coding apparatus according to the fourth embodiment. In the figure, an input terminal 1, a block determination circuit 2, an unequal division conversion coding circuit 3, an effective coefficient determination circuit 4, and output terminals 5 and 6 are the same as those in the first embodiment shown in FIG. Reference numeral 13 is a block information encoding circuit connected to the block determination circuit 2.

Next, the operation will be described. The coding apparatus according to the fourth embodiment is obtained by adding the block information coding circuit 13 that performs variable length coding to block structure information to the coding apparatus according to the first embodiment. As an encoding method of block structure information, for example, a signal is first divided into N-dimensional macroblocks of equal size, and the structure of block groups of various sizes included in each macroblock is examined, This can be realized by collecting and classifying those having the same block structure, obtaining the appearance probability, and performing variable length coding such as Huffman code or arithmetic code on each macroblock according to the appearance probability.

Further, by dividing the signal into a certain range, obtaining the appearance probability of the block structure within the range, and transmitting this appearance probability or information corresponding thereto,
It is also possible to perform variable length coding according to the appearance probability of the block structure within that range.

As described above, variable length coding such as Huffman coding or arithmetic coding is performed on the block structure information according to the appearance probability thereof, so that the code amount of the block structure information is reduced to near the entropy. be able to.

Example 5. Embodiment 5 of the present invention will be described below with reference to the drawings. FIG. 5 is a block circuit diagram of the transform coding apparatus according to the fifth embodiment. In the figure, an input terminal 1, a block determination circuit 2, an unequal division conversion coding circuit 3, an effective coefficient determination circuit 4, and output terminals 5 and 6 are the same as those in the first embodiment shown in FIG. Reference numeral 14 is a quantizer connected to the block determination circuit 2 and the effective coefficient determination circuit 4.

Next, the operation will be described. The coding apparatus according to the fifth embodiment is obtained by adding a quantizer 14 that adaptively quantizes transform coefficients to the coding apparatus according to the first embodiment. Based on the block structure information output from the block determination circuit 2, for example, in the image signal, it can be considered that only a low-frequency transform component is present in the image signal in a large block portion. Since the deterioration is easily detected in terms of visual characteristics, the deterioration is reduced by performing a fine quantization by reducing the quantization step width of the low-frequency transform coefficient.

On the contrary, in the small block portion, the electric power exists in the image signal over a wide conversion region from the low-frequency conversion component to the high-frequency conversion component, and it is difficult to detect the deterioration in all conversion regions due to the visual characteristics. However, even in such a case, deterioration is more likely to be detected in the low-frequency transform region than in the high-frequency transform region, so the quantization step width of the low-frequency transform coefficient corresponding to the small block corresponds to the large block. The low-frequency transform coefficient has a larger quantization step width, and the higher-frequency transform coefficient has a smaller quantization step width. By coarsely quantizing a small block in this way, it is possible to reduce the amount of codes to be transmitted without visually deteriorating.

There has been a quantization method for adaptively changing the quantization step width based on such a local property of the signal, but each conversion coefficient is quantized at which quantization step width. It was necessary to transmit the information only for adaptive quantization, which is whether or not it has been done. In Example 5,
Since the block structure information determines the quantization step width, it is not necessary to newly transmit information about what quantization step width each transform coefficient is quantized, and it is effective not only for adaptive quantization. Only the block structure information that can be used for the determination of the invalid transform coefficient needs to be transmitted, and more efficient encoding can be performed.

Example 6. Embodiment 6 of the present invention will be described below with reference to the drawings. FIG. 6 is a block circuit diagram of the transform decoding apparatus of the sixth embodiment, which corresponds to the inverse DCT operation circuit 911 in FIG. In the figure, 15 is an input terminal of the received block structure information, 16 is an input terminal of the received effective conversion coefficient, 17 is an invalid coefficient detection circuit connected to the input terminal 12, and 18 is an input terminal 16 and an invalid coefficient detection circuit. 17
Is a non-equal division conversion decoding circuit connected to the invalid coefficient substitution circuit 18, and 20 is an output terminal connected to the non-equal division conversion decoding circuit 19.

Next, the operation will be described. Example 1,
Alternatively, the transform coefficients output from the encoding device according to the second embodiment and received by the decoding device according to the sixth embodiment are only effective coefficients. When the effective coefficients are sequentially input from the input terminal 16, the conversion coefficient string does not include the invalid coefficient. Therefore, if the effective coefficients only are input to the unequal division conversion decoding circuit 19 as they are, The number of transform coefficients does not match, the correspondence with the decoding filter is lost, and incorrect transform decoding is performed.

In order to avoid this, first, the input terminal 15
On the basis of the block structure information input from, the invalid coefficient detection circuit 17 detects in which portion between the effective coefficient sequences the invalid coefficient should be inserted. The invalid coefficient substitution circuit 18 substitutes a constant at a position where the invalid coefficient detected by the invalid coefficient detection circuit 17 exists. here,
Since the invalid coefficient is assumed to have a conversion coefficient value of almost 0 in the first and second embodiments, the constant value to be substituted as the invalid coefficient in the sixth embodiment is 0.

In this way, the invalid coefficient detecting circuit 17 detects the position of the invalid coefficient, and the invalid coefficient substituting circuit 18 outputs the effective coefficient input from the input terminal 16 and the value obtained by substituting 0 as the value of the invalid coefficient. This enables the non-uniform division conversion decoding circuit 19 to perform band non-uniform division conversion decoding such as inverse wavelet conversion. The signal restored by the unequal division conversion decoding circuit 19 is
It is output from the output terminal 20.

Example 7. Embodiment 7 of the present invention will be described below with reference to the drawings. FIG. 7 is a block circuit diagram of the conversion decoding apparatus according to the seventh embodiment. In the figure, input terminals 15 and 1
6, invalid coefficient substitution circuit 18, unequal division conversion decoding circuit 1
9, the output terminal 20 is the same as that of the sixth embodiment shown in FIG. Reference numeral 21 is an invalid position detection circuit connected to the input terminals 15 and 16, and 22 is an input terminal 16 and an invalid position detection circuit 21.
And an inverse conversion determination circuit 2 connected to the inverse conversion determination circuit 2
Reference numeral 2 is an invalid coefficient decoding circuit, and reference numeral 24 is a multiplexer connected to the unequal division transform decoding circuit 19 and the invalid coefficient decoding circuit 23.

Next, the operation will be described. Similar to the sixth embodiment, only the effective coefficient is output from the first embodiment, the second embodiment or the like, and the decoding cannot be performed as it is. Therefore, in the seventh embodiment, two types of decoding are performed.
Either of two types of decoding methods is used depending on the condition satisfied by the conversion coefficient, and the judgment is made by the invalid position detection circuit 21 and the inverse conversion judgment circuit 22.

The invalid position detection circuit 21 detects the position of the invalid coefficient based on the block structure information input from the input terminal 15, similarly to the invalid coefficient detection circuit 17 of the sixth embodiment. Further, in this circuit, there may be a case where the effective coefficient input from the input terminal 16 takes a specific value (for example, 0).

In the inverse transform determination circuit 22, each transform coefficient is calculated based on the effective transform coefficient input from the input terminal 16 and the position information of the invalid coefficient or the null valid coefficient output from the invalid position detection circuit 21. Which of the two decoding methods is used for decoding is determined. When the invalid coefficient exists at a certain position (here, near the center of gravity of the decoding filter) of the decoding filter, the invalid coefficient decoding circuit 23 performs the decoding of that portion. Also, the effective coefficient that takes a certain specific value (for example, 0) is
Even when the decoding filter exists at a certain position (here, near the center of gravity of the decoding filter), the invalid coefficient decoding circuit 23 may perform the decoding of that portion. For decoding the part that does not satisfy the above conditions,
Similar to the decoding method of the sixth embodiment, the invalid coefficient substituting circuit 18 and the unequal division conversion decoding circuit 19 perform the decoding.

The first decoding method is performed for an invalid coefficient at a certain position of the decoding filter when it exists, but in this case, the invalid coefficient decoding circuit 2
3, the value after decoding by multiplying the transform coefficient and the decoding filter is always a constant and the multiplexer 24
Is output to. The presence of the invalid coefficient at a predetermined position means that the invalid coefficient exists around the center of gravity of the decoding filter.

The constant having a value after decoding is 0 in this case. Not only when the invalid coefficient exists near the center of gravity of the decoding filter, but also at a specific value (here, 0
Decoding using the invalid coefficient decoding circuit 23 may be performed even when there is an effective coefficient such as (4) that exists near the center of gravity of the decoding filter. When an invalid coefficient (which may include a valid coefficient having a value of 0) is present near the center of gravity of the decoding filter, the invalid coefficient decoding circuit 23 decodes the decoding filter with any value of the transform coefficient. By always setting the value of the multiplication with and to 0, especially when there is a large quantization error in the transform coefficient, the quantum that has a large power around the invalid coefficient or the effective coefficient of 0 value, It is possible to remove the influence of the high-frequency transform coefficient value that accompanies the digitization error. Therefore, for the part that is originally flat and has no high-frequency component, the influence of the surrounding high-frequency transform component is removed, and the decoding with a small error is performed. Can be converted.

Another decoding method is performed for an invalid coefficient (which may include a valid coefficient having a value of 0) when it is not near the center of gravity of the decoding filter. In this case, the decoding method is the same as that of the sixth embodiment, in which a constant (here, 0) in the invalid coefficient is substituted by the invalid coefficient substituting circuit 18, and the normal inverse transform is performed by the unequal division transform decoding circuit 19. Is done. The decoded signal obtained by the inverse transform is output to the multiplexer 24.

The multiplexer 24 multiplexes the decoded signals output from the unequal division transform decoding circuit 19 and the invalid coefficient decoding circuit 23, and the decoded signal is output from the output terminal 20.

Example 8. Embodiment 8 of the present invention will be described below with reference to the drawings. FIG. 8 is a block circuit diagram of the conversion decoding apparatus according to the eighth embodiment. In the figure, the input terminal 16, the invalid coefficient detection circuit 17, the invalid coefficient substitution circuit 18, and the non-uniform division conversion decoding circuit 19 are the same as those in the sixth embodiment shown in FIG. Reference numeral 25 is an input terminal for a motion vector code word encoded in the third embodiment, 26 is a block information decoding circuit connected to the input terminal 25, and 27 is an output terminal connected to the unequal division conversion decoding circuit 19. Is.

Next, the operation will be described. The transform decoding apparatus according to the eighth embodiment is obtained by adding the block information decoding circuit 26 to the transform coding apparatus according to the sixth embodiment. Example 3
The block structure information is not directly transmitted from the transform coding device such as. Therefore, it is necessary to restore the block structure information on the decoding device side. Therefore, since the motion vector codeword transmitted from the transform coding apparatus according to the third embodiment includes the block structure information,
Based on this motion vector code word, block structure information is restored. This can be realized by the block information decoding circuit 26 that performs an operation reverse to that of the motion vector coding circuit 11 of the third embodiment, that is, performs decoding. Then, the block structure decoding circuit 26 outputs the block structure information.

The block structure information output from the block information decoding circuit 26 is input to the invalid coefficient detection circuit 17. The subsequent operation is the same as that of the conversion decoding apparatus according to the sixth embodiment. Further, in the eighth embodiment, the block information decoding circuit 26 is added to the conversion decoding device of the sixth embodiment. However, even if the block information decoding circuit 26 is added to the conversion decoding device of the seventh embodiment, the embodiment is also provided. Decoding similar to that in 8 can be performed.

[0069]

As described above, according to the present invention, a signal is divided into blocks of various sizes according to the local property, and transform coding is also performed. Based on the size of the block representing, the transform coefficient obtained by transform coding is classified into an effective coefficient having important information and an invalid coefficient having no important information,
Since the invalid coefficient has a small amount of information, only the effective coefficient is outputted without outputting it. Therefore, there is an effect that the transmitted information amount can be reduced without deteriorating the decoded signal.

[Brief description of drawings]

FIG. 1 is a block circuit diagram of a transform coding apparatus according to a first embodiment of the present invention.

FIG. 2 is a block circuit diagram of a transform coding apparatus according to a second embodiment of the present invention.

FIG. 3 is a block circuit diagram of a transform coding apparatus according to a third embodiment of the present invention.

FIG. 4 is a block circuit diagram of a transform coding apparatus according to a fourth embodiment of the present invention.

FIG. 5 is a block circuit diagram of a transform coding apparatus according to a fifth embodiment of the present invention.

FIG. 6 is a block circuit diagram of a conversion decoding device according to a sixth embodiment of the present invention.

FIG. 7 is a block circuit diagram of a conversion decoding device according to a seventh embodiment of the present invention.

FIG. 8 is a block circuit diagram of a conversion decoding device according to an eighth embodiment of the present invention.

FIG. 9 is a block circuit diagram of a conventional transform coding apparatus.

[Fig. 10] Fig. 10 is a diagram for describing an example of band non-uniform division type transform coding.

[Explanation of symbols]

 2 block determination circuit 3 unequal division transform coding circuit 4 effective coefficient determination circuit 8 block size variable motion compensation predictive coding circuit 11 motion vector coding circuit 13 block information coding circuit 14 quantizer 17 invalid coefficient detection circuit 18 invalid Coefficient substitution circuit 19 Non-uniform division conversion decoding circuit 21 Invalid position detection circuit 22 Inverse transformation determination circuit 23 Invalid coefficient decoding circuit 24 Multiplexer 26 Block information decoding circuit

Claims (8)

[Claims] 1. An apparatus for encoding a signal using band non-uniform division conversion coding such as a wavelet transform or band non-uniform division sub-band coding in which a low frequency conversion base is longer than a high frequency conversion base. A block determination unit that divides an input signal into N-dimensional block structures of various sizes and outputs the block structure information, and a non-equal division transform coding unit that performs band non-equal division transform coding of the input signal. And based on the block structure information output from the block determination means, the transform coefficient output from the non-uniform partition transform coding means is divided into an effective coefficient and an invalid coefficient, and only the effective coefficient is output. A transform coding apparatus, comprising: an effective coefficient determining means. 2. An apparatus for encoding an image signal using band non-uniform division conversion coding such as wavelet transform or band non-uniform division sub-band coding in which a low frequency conversion base is longer than a high frequency conversion base. , The image signal is divided into N-dimensional blocks of various sizes, the motion-compensated prediction coding is performed on each block, and the motion-compensated prediction error signal, motion vector, and block structure information are output. Motion-compensated predictive coding means, non-equal-divided transform coding means for performing non-equal band conversion transform coding on the motion-compensated predictive error signal output from the block-size variable motion-compensated predictive coding means, and the block size. Based on the block structure information output from the variable motion compensation predictive coding means, the coefficients output from the non-uniform partition conversion coding means are effective coefficients and invalid coefficients. The transform coding apparatus is characterized by comprising: an effective coefficient determining means for outputting only an effective coefficient. 3. A motion vector output from this block size variable motion compensation predictive coding means is coded based on the block structure information output from the block size variable motion compensation predictive coding means. 3. The transform coding apparatus according to claim 2, further comprising motion vector coding means. 4. When the block structure information output from the block determining means according to claim 1 or the block size variable motion compensation predictive encoding means according to claim 2 is encoded and transmitted, the appearance probability of the block structure is determined. 3. The transform coding apparatus according to claim 1, further comprising block information coding means for variable length coding. 5. The effective coefficient determination according to claim 1 or 2 based on the block structure information output from the block determination means according to claim 1 or the block size variable motion compensation predictive coding means according to claim 2. 3. The transform coding apparatus according to claim 1, further comprising a quantizing means for changing a quantizing step width of each effective transform coefficient output from the means. 6. A transform decoding device for restoring an original signal from a signal encoded by the transform coding device according to claim 1, claim 2, claim 3, claim 4 or claim 5. The invalid coefficient detecting means for detecting which transform coefficient is not output as an invalid coefficient from the encoding device based on the block structure information output from the transform encoding device, and the invalid coefficient detecting means. Invalid coefficient assigning means for outputting the conversion coefficient determined as an invalid coefficient as a constant,
The non-equal band transform transform decoding such as the inverse wavelet transform or the non-equal subband decoding is performed on the transform coefficient output from the invalid coefficient substituting means and the received valid coefficient, and the non-equal output is output. An equal division conversion decoding means is provided, and the conversion decoding apparatus characterized by the above-mentioned. 7. The non-uniform division decoding means comprises means for replacing the value of the decoded signal with a predetermined constant value in the vicinity of the invalid coefficient or the effective coefficient having a specific value. The conversion decoding device according to claim 6. 8. The block information decoding means for outputting block structure information based on the motion vector output from the motion vector encoding means according to claim 3, is added. The described conversion decoding device.