JP3394619B2 - Image compression system - Google Patents

Description

Detailed Description of the Invention

[0001]

FIELD OF THE INVENTION This invention relates to digital image processing, and more particularly to digital image compression.

[0002]

2. Description of the Related Art As a standard compression method for still image,
JPEG (joint photographic e)
xpert group) compression method.

FIG. 9 is a block diagram showing a processing procedure of JPEG compression. The original image data Iuv is image data to be compressed and is represented by a spatial area. The original image data Iuv is 1 of the original image divided into 8 × 8 blocks.
It is a matrix that represents blocks. Matrix of original image data Iuv
The component of represents each pixel data of the original image.

Discrete Cosine Transform (hereinafter referred to as DCT)
The arithmetic processing circuit 51 applies a DC to the original image data Iuv.
The T operation is performed to generate the DCT coefficient Fuv. The DCT coefficient Fuv is information (spatial frequency component) represented in the frequency domain.

The DCT calculation processing circuit 51 performs DCT calculation on the 8 × 8 original image data Iuv. DC is performed by sandwiching the original image data Iuv between the transposed cosine coefficient matrix D ^t and the cosine coefficient matrix D and performing matrix operation.
The T coefficient Fuv is obtained.

F = D ^t ID (1) FIG. 10 is a diagram for explaining a process of obtaining a DCT coefficient Fuv by performing a DCT operation based on the original image data Iuv.

The original image data Iuv to be calculated is 8
It is a matrix of × 8, and the DCT coefficient Fuv which is the calculation result is also a matrix of 8 × 8. The matrix Fuv of DCT coefficients has a smaller number of rows and columns (towards the upper left of the matrix),
The coefficient of the low frequency component is represented. Conversely, the larger the number of rows and the number of columns (toward the lower right direction), the higher the frequency component.

When the 0th row (L0) of the original image data Iuv is input, the 0th row (L0) of the DCT coefficient Fuv can be calculated. The DCT calculation can be performed row by row. When the DCT calculation is performed, the DCT coefficient Fu
v is output in the order of raster scan.

FIG. 13 is a diagram showing the order in which raster scanning is performed in an 8 × 8 matrix. The raster scan starts from the 0th row (L0). 0th
The row (L0) is sequentially scanned from left to right. When the scan of the 0th row (L0) is completed, the scan of the 1st row (L1) is performed next. Thereafter, similarly, scanning is sequentially performed from the second row (L2) to the seventh row (L7).

In FIG. 9, the DCT coefficient Fuv obtained by the DCT operation is quantized in the quantization operation processing circuit 53 to obtain the quantized data Ruv. 8x
The DCT coefficient Fuv of 8 is divided by the quantization table Quv which changes depending on the frequency component, and finer quantization is performed as the frequency is lower, and coarser quantization is performed as the frequency is higher.

That is, the DCT coefficient Fuv is linearly quantized into Fuv / Quv by using the quantization table Quv having a finer step size as the components of row u and column v are smaller.

The quantized and rounded coefficient Ruv is represented by the following equation. Rounding means rounding to the nearest integer. Ruv = round (Fuv / Quv) (2) FIG. 11 is a diagram showing a matrix of the quantization table Quv. The quantization table Quv is composed of 8 rows and 8 columns and has a DC
The lower the frequency component with respect to the T coefficient Fuv, the smaller the value,
Performs fine quantization, and the higher the frequency component, the larger the value,
Coarse quantization is performed.

FIG. 12 shows an example of the coefficient Ruv obtained by performing the above-described quantization operation on the DCT coefficient Fuv for a general image block. By finely quantizing the low-frequency component and coarsely quantizing the high-frequency component, the quantized data Ruv of the high-frequency component becomes a small value. Generally, the quantized data Ru
Of v, high-frequency components (lower right part of the matrix) tend to have zero values.

In FIG. 9, the quantized data Ruv is run-length encoded and Huffman encoded in the encoding operation processing circuit 55 to generate compressed image data data.

The run-length coding can perform high compression on the data in which the value of 0 continues continuously. In the quantized data Ruv shown in FIG. 12, many 0s are gathered at the lower right of the matrix. By using this property, if the matrix Ruv of the quantized data is subjected to run-length coding by zigzag scanning instead of raster scanning, high compression can be performed.

FIG. 14 is a diagram showing the order in which zigzag scanning is performed in an 8 × 8 matrix. The zigzag scan scans from the upper left (0th) of the matrix to the lower right (63rd) in a zigzag shape. By performing a zigzag scan on the quantized data Ruv, it is possible to sequentially scan from the upper left of the matrix where the low frequency component is located to the lower right of the matrix where the high frequency component is located.

The quantized data Ruv (FIG. 12) has a property that high-frequency components (lower right of the matrix) are likely to have a value of 0 because of the above reasons. Therefore, in order to perform run-length coding, a raster scan (see FIG. Higher compression can be achieved by performing zigzag scanning (FIG. 14) than by performing 13).

In FIG. 9, a coding operation processing circuit 55 is provided.
Performs run-length coding and then Huffman coding to generate compressed image data data. The generated compressed image data data is stored in the storage medium.

As described above, in JPEG compression, both raster scan and zigzag scan are used as the data scanning method. In FIG. 9, the DCT arithmetic processing circuit 51 outputs the DCT coefficient Fuv in the raster scan. Then, the DCT coefficient Fuv is input to the quantization operation processing circuit 53 by zigzag scanning.

At this time, the data flow is converted from the raster scan to the zigzag scan using the block memory for storing the 8 × 8 DCT coefficient Fuv. DCT
The DCT coefficient Fuv output from the arithmetic processing circuit 51 is
It is written in the block memory in the order of raster scan. The DCT coefficient Fuv read from the block memory by the zigzag scan is input to the quantization operation processing circuit 53.

FIG. 15 is a diagram showing the timing of the DCT processing and the quantization processing when the block memory is used. The DCT processes 60 and 61 are processes performed by the DCT arithmetic processing circuit 51 (FIG. 9), and the quantization process 61 is
This is a process performed by the quantization operation processing circuit 53 (FIG. 9).

As described above, each process is performed in units of 8 × 8 block data. The DCT processing 60 for the n-th block performs DCT calculation, and the DCT coefficient Fu is stored in the block memory in the raster scan order.
Write v. After writing the last 63rd data of the 8 × 8 = 64 DCT coefficients Fuv, the quantization processing 61 of the nth block is started.

Quantization processing 61 is performed from 0th to 63th order.
The DCT coefficients Fuv up to the th are read out from the block memory in the order of zigzag scanning and processed. After the 63rd DCT coefficient Fuv is read, the DCT process 62 for the (n + 1) th block starts.

The DCT process 62 is performed in the order 63 from the 0th.
The DCT coefficients Fuv up to the th are written in the block memory in the order of raster scan. As described above, first, all 64 pieces of data are written in the block memory,
After that, reading of data is started. Then, after all 64 pieces of data have been read out, writing of data in the next block starts.

[0025]

When writing to the block memory by raster scanning and reading by zigzag scanning, the reading is performed after all the data in the block, which is the processing unit, has been written. I was going.

Generally, since the process of writing by raster scan includes DCT operation, it takes considerably longer time than the process of reading by zigzag scan (including quantization process), and the processing efficiency is high. Is bad, and the processing time is slowed down overall.

An object of the present invention is to provide an image compression system which can compress image data at high speed.

[0028]

An image compression system of the present invention comprises a block memory for storing two-dimensional block data, a raster address generating means for generating an address of the block memory in raster scan order, and a zigzag pattern. Zigzag address generating means for generating block memory addresses in the order of scanning, and DCT coefficients are generated by performing discrete cosine transform (hereinafter referred to as DCT), and DCT coefficients are generated at the block memory addresses generated by the raster address generating means. And a quantizing means for quantizing by reading the DCT coefficient from the address of the block memory generated by the zigzag address generating means, and the quantizing means is included in any row of the block memory. Detected that all the DCT coefficients CT processing unit to start writing on the row is read out the, and a control means for controlling the processing of the DCT processing means and raster address generator.

[0029]

The DCT processing means can write the DCT coefficient of the next DCT coefficient block in the block memory before the quantizing means reads all the blocks of the DCT coefficient stored in the block memory.

[0030]

1 is a block diagram showing the configuration of an image compression system according to a first embodiment of the present invention.

The image compression system of this embodiment is, for example,
A part of the JPEG compression processing is shown. The image compression system uses the DCT for the supplied original image data Iuv.
DCT processing is performed in the processing unit 1 to obtain the DCT coefficient Fuv.
Output r, and block memory 2 in raster scan order
Write to. The quantization unit 3 reads the DCT coefficient Fuvz from the block memory 2 in the order of zigzag scanning, performs quantization processing, and outputs quantized data Ruv.

The block memory 2 is a memory buffer capable of storing data for one block of 8 × 8. Hereinafter, a case where the block memory 2 is a single port memory will be described as an example.

In this embodiment, the speed of image compression processing is increased by controlling the timing of writing or reading of the block memory 2. Hereinafter, a method of controlling the timing will be described.

The DCT processing section 1 performs DCT processing on the supplied original image data Iuv and outputs the DCT coefficient Fuvr to the block memory 2 in the order of raster scanning. While the DCT coefficient Fuvr is input to the block memory 2, the raster address AR generated by the raster address generator 4 is input via the selector 6.

The DCT coefficient Fuvr output from the DCT processor 1 is a raster address AR in the block memory 2.
Writing is performed at the address specified by. The raster address AR is a memory address generated in the raster scan order (FIG. 13).

The selector 6 responds to the select signal SEL generated by the controller 7 to generate the raster address generator 4
The block memory 2 is supplied with either the raster address AR generated by the zigzag address generator or the zigzag address AZ generated by the zigzag address generator 5.

When the original image data Iuv of the first block (including all the 0th to 63rd data) is supplied, the controller 7 supplies the raster address A
A selection signal SEL for selecting R is supplied to the selector 6.

As shown in FIG. 2, in the DCT processing 10, when the DCT processing unit 1 writes the 63rd last DCT coefficient Fuvr in the block memory 2, the quantization processing 11
, The quantization unit 3 determines the 0th DCT coefficient Fuvz
To start reading. At this time, the controller 7 instructs the quantizer 3 to start processing and the zigzag address generator 5 to start generation of the zigzag address AZ. The zigzag address generator 5 starts generating the zigzag address AZ in order from the 0th data. Zigzag address AZ
Are memory addresses generated in the order of zigzag addresses.

In FIG. 1, the selector 7 receives the select signal SEL from the controller 7 and selects the zigzag address AZ generated by the zigzag generator 5. The zigzag address generator 5 uses the zigzag scan sequence (see FIG.
In 4), memory addresses are sequentially generated.

The zigzag address AZ is supplied to the block memory 2 via the selector 7. Quantizer 3
Reads the DCT coefficient Fuvz from the block memory 2 in the order of zigzag scanning.

As shown in FIG. 2, in the quantization processing 11, when the quantization unit 3 reads the 28th DCT coefficient Fuvz, the controller 7 starts the processing of the DCT processing unit 1 and causes the raster address generator 4 to operate. It is instructed to start generating the raster address AR. The 28th DCT coefficient Fuvz is the upper rightmost data as shown in FIG. 28
When the reading of the th data is completed, the 0th row (L0)
8 data can be written.

Upon receipt of the instruction from the controller 7, the raster address generator 4 starts generating raster addresses AR sequentially from the 0th data. After the reading of the 28th data of the quantization processing 11 is completed in the DCT processing unit 12 of FIG. 2, the DCT processing unit 1 determines the DC of the 0th row (0th to 7th) of the next block data.
Writing of the T coefficient Fuvr is started.

FIG. 3 is a diagram for explaining the timing of reading by the quantizer 3 and the timing of writing by the DCT processor 1 in the block memory 2. In FIG. 3A, when the quantizer 3 reads the 0th to 28th data from the block memory 2 by the zigzag scan, the 29th to 63rd data has not been read yet. The 0th to 28th data that have been read are not shown, and only the 29th to 63rd data that have not been read are shown.

When the 28th data is read, the 0th data is read.
Eight data in the row (L0) (0th, 1st in FIG. 14)
(5th, 5th, 6th, 14th, 15th, 27th, 28th) are all read, so D
The CT processing unit 1 can write the 0th row (L0) data (0th to 7th) of the next block data.

Then, in FIG. 3B, the quantizer 3
When the 29th to 42nd data are read from the block memory 2 by the zigzag scan, the 43rd to 63rd data have not been read yet. The read 0th to 42nd data are not shown, and only the 43rd to 63rd data that have not yet been read are shown.

When the 42nd data is read, the first
Eight data in the row (L1) (second and fourth in FIG. 14)
(7th, 7th, 13th, 16th, 26th, 29th, 42nd) are all read, so D
The CT processing unit 1 can write new data (8th to 15th) in the first row (L1).

Similarly, after the quantizing unit 3 reads out the 43rd data, the DCT processing unit 1 repeats the new second data.
The data in the row (L2) can be written. When the quantizer 3 reads the 53rd, 54th, 60th, 61st, and 63rd data, the third row (L
3), 4th row (L4), 5th row (L5), 6th row (L
6), the DCT processing unit 1 can write the data of the seventh row (L7).

The above switching between the read and write timings is controlled by the controller 7 in FIG. The controller 7 controls the processing start of the DCT processing unit 1 and the quantization unit 3 and controls the address generation of the raster address generator 4 and the zigzag address generator 5.

The above processing flow will be described with reference to FIG. The DCT process 10 is a process performed by the DCT processing unit 1 on the nth block. In the DCT processing 10, the DCT processing unit 1 calculates the 0th to 63rd DCT coefficients in the nth block and writes them in the block memory 2 in the raster scan order. When the 63rd DCT coefficient is written, the processing of the quantization processing 11 starts.

In the quantization processing 11, the quantization unit 3 is applied to the n-th DCT coefficient block calculated in the DCT processing 11.
Is a process performed by. In the quantization processing 11, the quantization unit 3 reads the DCT coefficients from the block memory 2 in order of zigzag scanning from the 0th, and performs quantization. When the reading of the 0th to 28th DCT coefficients is completed, the DCT process 12 starts.

In the DCT processing 12, the DCT processing unit 1 outputs n +
This is the process performed on the first block. In the DCT processing 12, the DCT processing unit 1 calculates the DCT coefficient of the 0th row (0th to 7th) in the (n + 1) th block and writes it in the block memory 2 in the order of raster scanning.

In the quantizing process 11, the quantizing unit 3 processes the 0th to 28th DCT coefficients of the n-th block in the zigzag scan order, and then processes the 29th to 42nd DCT coefficients. I do. When the processing (reading) of the 42nd DCT coefficient is completed, in the DCT processing 12, the DCT processing unit 1 calculates the DCT coefficient of the first row (8th to 15th) in the (n + 1) th block, and stores it in the block memory. 2 is written in raster scan order.

Similarly, in the quantization processing 11,
The 43rd, 53rd, 54th, 60th, and
When the 61st and 63rd DCT coefficients are read by the zigzag scan, respectively, in the DCT processing 12,
Second row (L2), third row (L3), fourth row shown in FIG.
The data of the row (L4), the fifth row (L5), the sixth row (L6), and the seventh row (L7) are written.

Although not shown, in the DCT process 12,
When the writing of the seventh row (L7) is completed, the quantization unit 3
Is the D + 1 of the (n + 1) th block, as in the quantization processing 11.
The reading of the CT coefficient is started.

In the quantization processing 11, the quantization unit 3 does not always wait for the DCT processing unit 1 in the DCT processing 10 to finish writing the seventh row (L7) including the 63rd data. It is not necessary to start reading. For example, when the DCT processing unit 1 finishes writing the fifth row (L5), the quantizing unit 3 may read up to the 20th DCT coefficient by zigzag scanning.

As described above, the quantization processing 11 and the DCT processing 12 can write or read data in the block memory 2 during the processing of each data block. In general, the DCT processing requires a longer processing time than the quantization processing. Therefore, by controlling the access timing to the block memory 2 as described above, the processing can be performed efficiently in time, and high speed processing is possible. It can perform various image compression.

The embodiment of the processing when compressing the original image data has been described above. Next, an embodiment when decompressing compressed compressed image data will be described. Figure 4
FIG. 6 is a block diagram showing a JPEG expansion process. JP
The EG decompression is a process for decompressing the image data by decompressing the compressed image data generated by the JPEG compression shown in FIG. 9 described above. JPEG extension is also J
Similar to PEG compression, processing is performed in units of 8 × 8 blocks.

The compressed image data data stored in the storage medium is Huffman-decoded and run-length-decoded in the decoding operation processing circuit 57, and the quantized data R is obtained.
uv is generated. Since the Huffman coding and the run length coding are lossless coding, the decoded quantized data Ruv is the quantized data R at the time of JPEG compression (FIG. 9).
Same as uv.

The quantized data Ruv is inversely quantized by the inverse quantization operation processing circuit 59 by multiplication with the quantization table Quv, and is returned to the DCT coefficient F'uv. As the quantization table Quv, the same quantization table as that used in JPEG compression in FIG. 11 is used.

F′uv = Ruv · Quv (3) The DCT coefficient F′uv is D generated at the time of JPEG compression.
The CT coefficient Fuv is represented by a DCT coefficient including a quantization error.

The DCT coefficient F'uv is the inverse DCT (hereinafter,
D in the opposite direction in the arithmetic processing circuit 61 (referred to as IDCT)
CT calculation is performed and converted into image data I′uv in the spatial domain. The IDCT operation processing circuit 61 sandwiches the DCT coefficient F′uv between the cosine coefficient matrix D and the transposed cosine coefficient matrix D ^t, and performs matrix operation to obtain decompressed image data I′uv.

I ′ = DF′D ^t (4) It is assumed that the decompressed image data I′uv includes DCT error and quantization error with respect to the original image data Iuv before JPEG compression. Restored.

Next, an example in which the present invention is applied to the above JPEG expansion will be described. FIG. 5 is a block diagram showing the configuration of the image expansion system according to the second embodiment of the present invention. The image expansion system of this embodiment is, for example, JPE.
A part of the G decompression process is shown.

The inverse quantizer 21 performs an inverse quantization process on the supplied quantized data Ruv to obtain the DCT coefficient F '.
Output uvz. The DCT coefficient F'uvz is written in the block memory 22 in the order of zigzag scanning.
The IDCT processing unit 23 reads the DCT coefficient F′uvr from the block memory 22 in the order of raster scan, and I
DCT processing is performed and decompressed image data I'uv is output.

As described above, the zigzag address generator 2
4 generates the zigzag address AZ, and the raster address generator 25 generates the raster address AR. Selector 2
6 is a zigzag address AZ according to the selection signal SEL.
Alternatively, the raster address AR is selected and supplied to the block memory 22.

In the previous JPEG compression embodiment, data is written in the block memory 22 by raster scan and then read by zigzag scan. However, in the case of JPEG decompression, data is written in the block memory 22 by zigzag scan and Read by raster scan.

Hereinafter, the timing at which the zigzag address generator 24 generates the zigzag address AZ, the timing at which the raster address generator 25 generates the raster address AR, and the timing at which the inverse quantization unit 21 and the IDCT processing unit 23 start processing. Will be explained.

FIG. 6 shows the timing at which the inverse quantization unit 21 writes to the block memory 22 and the IDCT processing unit 2.
FIG. 3 is a diagram for explaining the timing of reading by 3; FIG. 6A is a schematic diagram of the block memory 22 when the dequantization unit 21 writes the 0th to 28th data in the block memory 22 by the zigzag scan. Only the written 0th to 28th data are shown.

When the 28th data is written, the 0th data is written.
Since all eight data (0th, 1st, 5th, 6th, 14th, 15th, 27th, 28th data in the zigzag scan) of the row (L0) have been written, the IDCT processing is performed. The unit 23 can read the data of the 0th row (L0) stored in the block memory 22.

In FIG. 6B, subsequently, the block memory 22 when the inverse quantizer 21 writes the 29th to 42nd data in the block memory 22 by the zigzag scan.
FIG. Only the 0th to 42nd data already written are shown.

When the 42nd data is written, the first
Since all the eight data in the row (L1) (the second, fourth, seventh, thirteenth, sixteenth, twenty-sixth, twenty-ninth, and thirty-second data in the zigzag scan) have been written, the IDCT processing is performed. The unit 23 can read the data of the first row (L1) stored in the block memory 22.

Thereafter, in the same manner, the inverse quantizing unit 21 uses 43
After writing the second data, the IDCT processing unit 23 can read the data in the second row (L2). The inverse quantizer 21 uses the 53rd, 54th, 60th, 61st,
When the 63rd data is written, the 3rd data
The data of the row (L3), the fourth row (L4), the fifth row (L5), the sixth row (L6), and the seventh row (L7) are transferred to the IDCT processing unit 2.
3 can read.

The controller 27 shown in FIG. 5 controls the switching of the write and read timings. The controller 27 includes an inverse quantization unit 21 and an IDCT processing unit 23.
Control the start of processing and the address generation of the zigzag address generator 24 and the raster address generator 25.

The case of DCT processing and quantization processing in the case of JPEG compression and decompression has been described above. Next, not only the JPEG method but also a case of writing to the block memory by raster scan and reading by raster scan will be described as an example.

FIG. 7 is a block diagram showing the arrangement of an image compression system according to the third embodiment of the present invention. The image compression system of this embodiment is similar to the first embodiment shown in FIG. 1 except that the line memory 32 is used instead of the block memory 2.
To use. The line memory 32 sequentially accesses the data in one line from the beginning to the end by raster scanning.

The DCT processing section 31 performs DCT processing on the supplied original image data Iuv to obtain the DCT coefficient F
uv1 is output, and line data is written to the line memory 32 by raster scanning. The quantizing unit 33 receives the DCT coefficient F from the line memory 32 in the order of raster scanning.
Read the uv2 line data, perform the quantization process,
The quantized data Ruv is output.

The write address generator 34 generates a write address A1 to be written in the line memory 32, and the read address generator 35 is used in the line memory 32.
A read address A2 for reading from is generated.
The selector 36 supplies the write address A1 or the read address A2 to the line memory 32 according to the selection signal SEL.

The DCT processing section 31 performs DCT processing on the supplied original image data Iuv, and the DCT coefficient Fuv1 of 1 is stored in the line memory 32 in the order of raster scanning.
Write lines. The quantizer 33 reads the DCT coefficient Fuv2 from the line memory 32 in the order of raster scan. DCT coefficient F read in raster scan order
uv2 is quantized and quantized data Ruv is output.

Hereinafter, the timing at which the write address generator 34 generates the write address A1, the timing at which the read address generator 35 generates the read address A2, and the timing at which the DCT processing section 31 and the quantizing section 33 start processing will be described. explain.

The DCT processing section 31 uses the original image data Iuv
Is subjected to DCT processing, and the DCT coefficient Fu of the 0th line (L0)
v1 is written in the line memory 32. Line 0 (L
When the writing of the DCT coefficient Fuv1 of 0) is completed, the quantization unit 33 reads the DCT coefficient Fuv2 of the 0th row (L0) from the line memory 32, performs the quantization process, and generates the quantized data Ruv.

Similarly, the DCT processor 31 performs the raster scan in the first row (L1), the second row (L2), the third row (L3), the fourth row (L4), and the fifth row (L5). , The sixth row (L6) and the seventh row (L7) are respectively written, the quantizer 33 writes the respective line data and then the first row (L1), the second row (L2), The data of the third row (L3), the fourth row (L4), the fifth row (L5), the sixth row (L6), and the seventh row (L7) are read and processed.

Although the addresses A1 and A2 have been described as being generated by the write address generator 34 and the read address generator 35, they may be generated by one address generator.

FIG. 8 is a block diagram showing the arrangement of an image expansion system according to the fourth embodiment of the present invention. The decompression / compression system of this embodiment is a system that decompresses the quantized data Ruv to generate decompressed image data I′uv, contrary to the image compression system shown in FIG.

Hereinafter, the timing at which the write address generator 44 generates the write address A1, the timing at which the read address generator 45 generates the read address A2, and the timing at which the dequantization unit 41 and the IDCT processing unit 43 start processing. Will be explained.

The inverse quantization unit 41 inversely quantizes the quantized data Ruv, and the DCT coefficient F'uv of the 0th line (L0).
2 is written in the line memory 42. Line 0 (L0)
When the DCT coefficient F′uv2 of No. 1 is written, the IDCT processing unit 43 reads the D of the 0th row (L0) from the line memory 42.
Read the CT coefficient F'uv1 and perform IDCT processing,
The decompressed image data I′uv is generated.

Similarly, the inverse quantizer 41 performs the raster scan on the first row (L1), the second row (L2) and the third row (L).
3), 4th row (L4), 5th row (L5), 6th row (L
6), writing the data of the 7th row (L7),
The IDCT processing unit 43 writes the first line data (L1), the second data line (L2), the third data line (L3), the fourth data line (L4), the fifth data line (L5), and the The data in the 6th row (L6) and the 7th row (L7) are read and processed.

The DCT operation is performed in one block (for example, 8 ×).
Since 8) is the unit of processing, in the conventional image compression or decompression processing, one block is naturally stored in the block memory. However, it is also possible to store one line as a unit by using the line memory as in this embodiment. By storing the data line by line, the waiting time of the processing can be reduced, and the speed of image compression or expansion can be increased.

The addresses A1 and A2 may be generated by one address generator. Further, since the line memory only needs to have a function of delaying the input data by a predetermined time and outputting the data, a first-in first-out circuit (FIFO) or the like may be used instead.

As described above, by controlling the access timing of the block memory or the line memory,
The image compression or decompression processing speed can be increased without increasing the memory capacity.

The block memory or the line memory can be applied not only to the single port memory but also to the dual port memory. In this case, writing and reading can be performed from separate ports, increasing the degree of freedom in timing adjustment.

The present invention has been described above with reference to the embodiments.
The present invention is not limited to these. For example, it will be apparent to those skilled in the art that various modifications, improvements, combinations, and the like can be made.

[0092]

As described above, according to the present invention,
The DCT processing means, before the quantizing means reads all the blocks of DCT coefficients stored in the block memory,
Since the DCT coefficient of the next DCT coefficient block can be written in the block memory, the image compression processing can be performed at high speed.

[Brief description of drawings]

FIG. 1 is a block diagram showing a configuration of an image compression system according to a first embodiment of the present invention.

FIG. 2 is a diagram showing temporal timing of DCT processing and quantization processing.

FIG. 3 is a diagram for explaining a timing read by a quantizer and a timing written by a DCT processor in a block memory.

FIG. 4 is a block diagram showing a JPEG expansion process.

FIG. 5 is a block diagram showing a configuration of an image expansion system according to a second embodiment of the present invention.

FIG. 6 is a diagram for explaining the timing of writing by the inverse quantizer to the block memory and the timing of reading by the IDCT processor.

FIG. 7 is a block diagram showing a configuration of an image compression system according to a third embodiment of the present invention.

FIG. 8 is a block diagram showing a configuration of an image expansion system according to a fourth embodiment of the present invention.

FIG. 9 is a block diagram showing a processing procedure of JPEG compression.

FIG. 10 is a diagram for explaining DCT calculation.

FIG. 11 is a diagram showing a matrix of a quantization table Quv.

FIG. 12 is a diagram showing a matrix of coefficients Ruv obtained by performing a quantization operation on DCT coefficients Fuv for a general image block.

FIG. 13 is a diagram showing the order of raster scanning in an 8 × 8 matrix.

FIG. 14 is a diagram showing the order of zigzag scanning in an 8 × 8 matrix.

FIG. 15 is a diagram showing temporal timing of performing DCT processing and quantization processing according to a conventional technique.

[Explanation of symbols]

1,31 Discrete Cosine Transform (DCT) Processing Unit 23, 43 Inverse Discrete Cosine Transform (IDCT) Processing Unit 2,22 block memory 32, 42 line memory 3,33 Quantizer 21,41 Dequantizer 4,25 raster address generator 5,24 zigzag address generator 34,44 write address generator 35, 45 read address generator 6,26,36,46 Selector 7, 27, 37, 47 controller 10, 12, 60, 62 DCT processing 11,61 Quantization processing 51 DCT arithmetic processing circuit 53 Quantization arithmetic processing circuit 55 Encoding operation processing circuit 57 Decoding operation processing circuit 59 Inverse quantization arithmetic processing circuit 61 IDCT arithmetic processing circuit

─────────────────────────────────────────────────── ─── Continuation of front page (58) Fields surveyed (Int.Cl. ⁷ , DB name) H04N ⁷ /24-7/68 H04N 1/41-1/419 G06T 1/60

Claims (6)

(57) [Claims] 1. A block memory for storing two-dimensional block data, a raster address generating means for generating an address of the block memory in a raster scan order, and an address of the block memory in a zigzag scan order. Zigzag address generating means for performing DCT by performing discrete cosine transform (hereinafter referred to as DCT)
DCT processing means for generating a T coefficient and writing the DCT coefficient to the address of the block memory generated by the raster address generating means, and a DCT coefficient from the address of the block memory generated by the zigzag address generating means. Read
Quantizing means for performing quantization, and detecting that the quantizing means has read out all the DCT coefficients contained in any row of the block memory, and the DCT processing means writes in the read row. Image compression system having DCT processing means and control means for controlling the processing of the raster address generating means. 2. A run length coding means for run length coding the DCT coefficient quantized by the quantizing means, and a Huffman coding for the code coded by the run length coding means. The image compression system according to claim 1, further comprising: 3. An image decompression system for decompressing image compression data generated by performing processing including DCT and quantization, comprising a block memory for storing two-dimensional block data, and a zigzag scan. Zigzag address generating means for generating the address of the block memory in the order of, raster address generating means for generating the address of the block memory in the order of raster scan, and dequantizing the image compressed data to generate the DCT coefficient Dequantizing means for writing the DCT coefficient to the address of the block memory generated by the zigzag address generating means, and reading the DCT coefficient from the address of the block memory generated by the raster address generating means, Inverse discrete cosine transform (hereinafter referred to as IDCT ID to perform
Detecting that the CT processing means and the inverse quantization means have written all the DCT coefficients included in any row of the block memory,
An image decompression system having an IDCT processing means and a control means for controlling the processing of the raster address generating means so that the IDCT processing means starts reading the written line. 4. A Huffman decoding means for Huffman decoding the compressed image data, and a run length decoding means for run length decoding the data decoded by the Huffman decoding means. 4. The image decompression system according to claim 3, wherein the inverse quantization means inversely quantizes the data decoded by the run length decoding means. 5. An image compression method using an image compression system having a block memory for storing two-dimensional block data, wherein DCT is performed to generate DCT coefficients, and the DCT coefficients are stored in the block memory in raster scan order. It includes a DCT processing step of writing the DCT coefficient and a quantization step of reading the DCT coefficient from the block memory in a zigzag scan order to perform quantization, and the DCT processing step includes any one of the block memories in the quantization step. An image compression method that detects that all the DCT coefficients included in the row have been read and starts writing in the read row. 6. An image decompression system having a block memory for storing two-dimensional block data,
An image decompression method for decompressing image compression data generated by performing DCT and quantization, wherein the image compression data is inversely quantized to generate DCT coefficients, and the DCT coefficients are stored in a block memory in a zigzag scan order. And an IDCT processing step of performing IDCT by reading DCT coefficients from the block memory in the order of raster scan, and the IDCT processing step is performed in any of the block memories in the dequantization step. An image expansion method that detects that all DCT coefficients have been written to a row and starts reading of the written row.