JPH0513434B2 - - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION Field of Industrial Application The present invention relates to a block encoding device used for highly efficient encoding of images and the like.

2. Description of the Related Art Coding that removes redundancy in image information and improves transmission efficiency is called high-efficiency coding. As a high-efficiency encoding method, there is a block encoding method in which a specific number of image samples are collected into blocks, and compression is performed within the blocks. This method includes an orthogonal transformation method in which sample values within a block are orthogonally transformed and quantized, and a vector quantization method in which each block is directly quantized. In either method, compression is performed by utilizing the strong correlation between adjacent samples of an image, so it is preferable that the samples within a block are as close as possible.

Here, a method of dividing one block into a block consisting of eight samples will be explained.
Figure 6a shows one-dimensional blocking, where the distance between the samples at both ends is quite large. On the other hand, FIG. 6b shows two-dimensional blocking, in which the distance between the specimens at both ends is considerably shortened by utilizing the two-dimensional continuous nature of the image. Furthermore, for images that have continuity in the time axis direction, such as TV signals, by applying three-dimensional blocking as shown in FIG. 6c, the distance between the samples at both ends can be further shortened.

By block encoding image information two-dimensionally or three-dimensionally in this way, greater compression than one-dimensional block encoding is possible.

Problems to be Solved by the Invention As described above, two-dimensional or three-dimensional encoding is generally used for highly efficient encoding of image information. However, in two-dimensional blocking, if the image changes significantly within a block, the correlation between samples becomes small, resulting in large distortions.

On the other hand, in three-dimensional blocking, if the image does not vary in the time-axis direction, the correlation in the time-axis direction is extremely high, and greater compression is possible than in two-dimensional blocking. However, in images with large movements in the time axis direction, there is almost no correlation in the time axis direction, resulting in a problem of large distortion.

SUMMARY OF THE INVENTION In view of the above, an object of the present invention is to provide a block encoding device that improves the above-mentioned drawbacks.

Means for Solving the Problems The present invention divides sample points of an image that are continuous in three dimensions including the time axis direction in a three-dimensional space to create large blocks each consisting of N sample points. and K types of second blocking means for further dividing the large block into small blocks, and K types obtained by each of the K types of second blocking means. block encoding means for generating a set of K types of block code words by block encoding each set of small blocks; a selection means for selecting a set of block codewords, and further transmits the set of block codewords selected by the selection means to a receiver receiving the transmitted set of block codewords. The block encoding device is characterized in that it has identification means for notifying whether a set of block code words generated from the blocking means has been selected.

Effect: With the above-described configuration, the present invention selects a small block that is long in the time axis direction for a large block that has a large correlation in the time axis direction, and selects a small block that is short in the time axis direction for a large block that has a small correlation in the time axis direction. This makes it possible to perform high-efficiency encoding with high quality and high compression rate for any image.

Embodiment First, as a first embodiment, an apparatus will be described which blocks images that are correlated also in the temporal direction and compresses the blocks by orthogonal transform encoding.

FIG. 1a shows a large block generated by the first blocking of the first embodiment, and this block has a total of 16 blocks, 2 in the vertical direction, 4 in the horizontal direction, and 2 in the time axis direction. This is a block consisting of (=N) samples. Furthermore, FIGS. 1b and 1c show how the large block shown in FIG. 1a is converted into a second block. Figure 1b shows a large block divided into 2-dimensional blocks (small blocks) consisting of 8 samples, and Figure 1c shows a large block divided into 3-dimensional blocks (small blocks) consisting of 8 samples. It shows what is happening.

Next, FIG. 2 shows a block diagram of the apparatus of the first embodiment. In FIG. 2, 1 is an input part of this device, 2 is a first blocking part, 3 is a two-dimensional small block part, 4 is a three-dimensional small block part, 5 and 6 are orthogonal transform coding parts, 7 and Reference numeral 8 indicates a distortion measuring section, 9 indicates an output selection section, and 10 indicates an output section of this device.

First, the sample value of 8 bits per sample inputted from the input section 1 of FIG. 2 is converted into a large block in the first blocking section 2, and a large block as shown in FIG. 1a is generated. The generated large block is then input to a two-dimensional small block section 3 and a three-dimensional small block section 4, and is divided into small blocks as shown in FIGS. 1b and 1c, respectively. The two two-dimensional small blocks obtained in the two-dimensional small block conversion section are each orthogonally transformed encoded in the orthogonal transform encoding section 5, and each is converted into a 32-bit code word, resulting in a total of 64-bit code word. . However, among these 64 bits, at least one of the first 2 bits is encoded to take a non-zero value.

Similarly, the two three-dimensional small blocks obtained in the three-dimensional small block conversion section are also orthogonally transformed encoded in the orthogonal transform encoding section 6, and each is converted into a code word of 31 bits. and the total obtained by this
Add 2 bits of 0 to the beginning of the 62 bits code word.
Constructs a 64bits codeword. For the two 64-bit code words obtained in the above manner, the magnitude of distortion due to orthogonal transform encoding is calculated by the distortion measuring sections 7 and 8, respectively. The magnitudes of the two distortions thus obtained are compared in the output selection section 9, and the codeword with the smaller distortion is selected, and the output section 1
Output to 0.

In this way, 128 bits (8 bits x 16 samples) of image information is encoded into 64 bits with high efficiency. Also, during decoding, if the first 2 bits of a 64-bit code word are both 0, it indicates that it has been converted into a 3-dimensional small block.
If this is not the case, the data can be uniquely decoded because it is known that the data has been converted into two-dimensional small blocks.

With this device, three-dimensional block encoding is applied to images with high correlation in the time axis direction, such as still images, and conversely, two-dimensional block encoding is applied to images with small correlation in the temporal axis direction, such as moving images. Therefore, it is possible to perform highly efficient encoding with small distortion and high compression ratio for images with any correlation in the time axis direction. Furthermore, in this embodiment, since there is no need to separately transmit information as to which block code has been selected, the compression rate does not deteriorate due to the transmission of selection information.

Next, a second embodiment using vector quantization will be described.

FIG. 3a shows a large block of this embodiment, and this block is a block consisting of a total of 64 sample points, each consisting of four samples in the vertical, horizontal, and time axis directions. Also, since each sample is represented by 8 bits, one large block consists of 512 bits.

Next, the appearance of the four types of small blocks obtained by forming the four types of second blocks is shown in Figure 2 b, c,
Shown in d and e. b is a small block with a length of 1 in the time axis direction, c and d are small blocks with a length of 2 in the time axis direction, and e is a small block with a length of 4 in the time axis direction. Small block, c
These are called small-type blocks, d-type blocks, and e-type small blocks. These small blocks are vector quantized for each small block, and as a result, b is 40 bits x 4 = 160 bits, and c and d are 40 bits per large block.
32bits×4=128bits, e is compressed to 24bits×4=96bits. In this embodiment, one code word is selected from the four small block results and transmitted together with selection information indicating which code has been selected.

FIG. 4 is a block diagram of the apparatus of this embodiment.
11 in Figure 4 is the large block input part, 12 to 15
are small block parts from b type to e type, 16 to 19 are vector quantizers, and 20 to 23 are
is the distortion measurement part, 24 is the output selection part, 25, 26
27 to 29 are buffers, 30 is an output selection portion, and 31 is an output switching portion. 32 indicates the output part of the code word.

8bits input from large block input part 11
The sample values of ×64=512 bits are input to four small block forming parts from the b-type small block forming part 12 to the e-type small block forming part 15, respectively, and the resulting small blocks are input to the vector quantizers 16 to 15. 1
9. As a result, the code word obtained from the b-type small block is 160 bits, the c-type and d-type is 128 bits, and the e-type is 96 bits.

Next, the distortion measurement parts 20 to 23 calculate the magnitude of distortion caused by each encoding, and from the results, the output selection part 24 calculates the code word obtained from c-type or d-type small block formation. The one with smaller distortion is selected. At the same time, the improvement level 1 measurement part 25 is
The result of subtracting the magnitude of distortion determined by the distortion measurement section 20 from the magnitude of distortion due to the code word selected by the output selection section 24 is output. Similarly, the improvement level 2 measurement section 26 also outputs the result of subtracting the magnitude of distortion determined by the distortion measurement section 23 from the magnitude of distortion due to the code word selected by the output selection section 24. .

The three types of code words obtained in the above manner are stored in buffers 27-29, respectively. In this embodiment, it is assumed that code words corresponding to 360 large blocks are stored. In addition, in the output selection section 30,
Improvement level 1 measurement part 25 and improvement level 2 measurement part 2
The outputs of 6 are arranged in descending order as shown in FIG.
Here, if the degree of improvement 1 or 2 is positive, it means that the code word obtained from the b-type or e-type small block formation has smaller distortion than the result of the output selection part 24. If both improvement degrees 1 and 2 obtained from the same block are positive, the larger improvement degree is retained and the smaller improvement degree is removed from the table.
Next, as shown in Figure 5 1 and 2, the improvement level 2 for positive values
If the number of positive values is equal to or greater than the number of positive improvement degrees of 1, b
A code word obtained from a small block of type e is selected, and a code word obtained from a small block of e type is selected for a block having a positive value of improvement degree 2. In addition, when the number of positive improvement degrees 1 is larger as shown in FIG. 5, the code word obtained from the b-type and e-type blocks is Select each. For blocks other than these, the code word obtained from the c-type or d-type block obtained as a result of the output selection section 24 is selected. With the above output selection, the e-type block (code length =
96bits) is selected from the b-type block (code length =
160bits) is equal to or greater than the number chosen.
Also, since the code word selected from the c-type and d-type blocks is 128 bits, the overall average code length is
It will be 128bits or less.

The outputs of the buffers 27 to 29 are output from the output switching section 31.
The output part 3 is switched as shown above, and the output selection information (2 bits because 4 types can be selected) is added to the output part 3.
Output to 2.

This embodiment is a device that compresses 512 bits to 130 bits, about 1/4, and can efficiently compress both moving images and still images.

Effects of the Invention As explained above, the present invention is a three-dimensional block encoding device that is applicable to both images with a high correlation in the time axis direction and images with a small correlation in the time axis direction, and is applicable to both moving images and still images. This method performs high-efficiency encoding with high quality and high compression rate.

[Brief explanation of drawings]

Fig. 1 is a sample value block diagram for explaining the first embodiment of the present invention, Fig. 2 is a block diagram of the same embodiment, and Fig. 3 is a sample value block diagram for explaining the second embodiment. FIG. 4 is a block diagram of the same embodiment, FIG. 5 is an explanatory diagram of block selection in the same embodiment, and FIG. 6 is an explanatory diagram of block formation. 2...First block formation, 3, 4...Second block formation, 5, 6...Orthogonal transform coding, 7, 8...Distortion measurement, 9...Output selection, 16-19...Vector quantization , 25, 26...improvement measurement, 27-29
...Batsuhua.

Claims (1)

[Scope of Claims] 1. A first method that divides sample points of an image having continuity in three dimensions including the time axis direction in a three-dimensional space to form a large block consisting of N sample points each.
K has a block forming means, and a second block forming means for further dividing the large block into small blocks.
a block encoding means for generating a set of K kinds of block code words by respectively block encoding a set of K kinds of small blocks obtained by each of the K kinds of second blocking means; It has a selection means for selecting one set of block codewords from a set of K types of block codewords obtained by the encoding means, and further includes a selection means for selecting one set of block codewords from among the set of K types of block codewords obtained by the encoding means, and transmit the set,
1. A block encoding device comprising: identification means for informing a recipient of a transmitted set of block codewords which second blocking means has selected a set of block codewords generated from the second blocking means. 2 When K=2, one second blocking means divides a large block into small blocks in a two-dimensional space, and another second blocking means divides a large block into small blocks in a three-dimensional space. The block encoding device according to claim 1, characterized in that the block encoding device performs division. 3. Claim 1 or claim 3, characterized in that the selection means selects a set of block code words that minimize distortion from the original sample value after decoding, from K types of block code word sets. 2. Block encoding device according to item 2. 4. A patent claim characterized in that when there are K types of block codes with different code lengths, the selection means selects a set of block code words so that the amount of transmission is always constant within a certain range. A block encoding device according to the first, second, or third range. 5. Block encoding according to claim 1, 2, 3, or 4, characterized in that the identification means enables identification by transmitting an identification signal separately from the block code word. Device. 6. When the block encoding means has the characteristic that K types of block codes do not have the same code word, the identifying means uniquely identifies the block code word by examining the transmitted block code word. A block encoding device according to claim 1, 2, 3, or 4. 7. Claims 1 and 2 characterized in that the block encoding means is orthogonal transform encoding.
6. A block encoding device according to item 1, 3, 4, 5, or 6. 8 Claims 1 and 2 characterized in that the block encoding means is vector quantization.
6. A block encoding device according to item 1, 3, 4, 5, or 6.