JPH09205552A - Image coding method and device therefor - Google Patents

Abstract

PROBLEM TO BE SOLVED: To increase the compression ratio of a natural still image. SOLUTION: In the case of coding, input image data are once reduced by an image reduction device 101 and the resulting data are magnified again by an image magnification device 102 and a difference with respect to the input image data is obtained by a difference device 103. Then difference image data are subjected to discrete cosine transformation and quantization in the unit of blocks and a Huffman coder 43 generates coded data only on a block whose all coefficients are not zero. Furthermore, an interleave device 104 generates an interleave table whose contents are flags to identify blocks subject to Huffman coding and blocks not Huffman-coded, and the table, the coded data and the reduced image data being compressed data are outputted. In the case of decoding, a coding data insert device 106 generates data whose picture element value is constant with respect to the blocks not coded and difference image data are reproduced by applying inverse processing to the coding to the data or the coded data subjected to Huffman coding and the difference image data and the magnified image data of the reduced image data are added by an adder 107.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a still image encoding method and apparatus, and more particularly to a natural still image compression method capable of attaining a high compression rate by partially reducing information of an original image. The present invention relates to a suitable image encoding method and apparatus.

[0002]

2. Description of the Related Art Image data encoding methods used for image communication and image storage include a "reversible method" capable of decoding a complete original image, and a partial reduction of information of the original image at the time of encoding. Although there is a "lossy method" that causes deterioration when decoding, the present invention relates to a latter still image coding method by the "lossy method" and its apparatus.

JP is used as a method for encoding still image data.
EG (Joint Photo Graphic Ex
perts Groups, ITU-T. 81) is often used. FIG. 4 is a block diagram showing a schematic configuration of a JPEG baseline type processing system, and FIG. 5 is a flowchart of this processing. First, when encoding,
The digitized image data is input and divided into blocks (hereinafter, this block is referred to as 8 × 8 pixels) (step 501).

Next, the blocks are taken out one by one (steps 502 and 508), and the processing of steps 503 to 506 is repeated until all the blocks are processed (step 507). In the processing for each block, the discrete cosine transformer 41 performs a discrete cosine transform (DCT:
Descrete Cosine Transform
m) (step 503) and the resulting DC
Each of the T coefficients is sub-quantized by the quantizer 42 (step 504). The value of the quantized coefficient is 0
Alternatively, only the coefficients excluding the DCT coefficient having a value close to 0 are entropy-coded by the Huffman encoder 43 as effective coefficients (step 505) and sent as compression encoded data to the communication channel or the storage medium 44 (step). 506).

Next, at the time of decoding, the compression-coded data is fetched from the communication channel or the storage medium 44 (step 509), and the coded data for each block is sequentially fetched (steps 510, 516), and step 511. The process of steps ˜514 is repeated until all blocks are processed (step 515). In the process for each block, the Huffman decoder 45 performs entropy decoding (step 511) and the inverse quantizer 46
Inverse quantization is performed by (step 512). Further, the inverse discrete cosine transformer 47 performs the inverse discrete cosine transform (step 513), and the expanded image data is output (step 514).

[0006]

According to the above JPEG method, it is possible to compress image data of about 1/50. However, if the compression is further performed, the restored image deteriorates severely, which is practically impossible. An object of the present invention is to solve the problem that the compression limit of natural image data by the lossy method of the related art is about 1/50, and an image of practical image quality can be obtained even if compressed at a higher compression rate. An object of the present invention is to provide an image encoding method that can be restored and an apparatus thereof.

[0007]

The present invention generates reduced image data from input original image data, enlarges the reduced image data to the same size as the original image data, and generates enlarged image data. , Further generate difference image data representing a difference between the enlarged image data and the original image data, divide the difference image data into a plurality of blocks, perform orthogonal transformation and quantization on each block, A table in which a flag having "0" for the block in which all the quantized data is 0 and "1" for the block in which all the data are not 0 is given to each of the blocks. Is generated, and the quantized data is encoded only for the block in which the flag is “1” to generate the encoded data. Thus, the reduced image data, the encoded data and the And the reduced image data output by the encoding process and the table are input, and the block whose flag is “1” is the block concerned. For the corresponding encoded data, decoding is performed on the encoded data when the pixel value is constant for the block whose flag is "0", and the decoded data is inverted. The decoded difference image data corresponding to the difference image data is generated by quantizing and the output is inversely orthogonally transformed, and the input reduced image data is enlarged to the same size as the original image and then the decoded difference image is generated. Disclosed is an image encoding method characterized by performing a decoding process of adding reproduced pixel data of an original image by adding pixel values of image data.

The present invention further includes an image reducer for generating reduced image data from input original image data, and an enlarged image for expanding the reduced image data to the size of the original image data. An image enlarger, a difference device for generating difference image data by taking the difference between the enlarged image generated by the image enlarger and the original image data, and the differential cosine transform of the difference image data in block units DCT
A discrete cosine transformer for generating coefficients, said DC
A quantizer for quantizing T coefficients, and "0" for a block in which the quantized DCT coefficients are all 0,
A decimator for generating a table in which "1" is given as the value of the flag corresponding to the block which is not all 0, and the quantized DCT of the block in which the flag is "1" A Huffman encoder for entropy-encoding the coefficient and a Huffman encoder are provided in the encoding means, and insert-encoded data whose pixel value is constant is generated for a block whose flag is "0". For thinning decision and inserter, and the block whose flag is "0", the inserted coded data from the thinning decision and inserter is encoded for the block whose flag is "1". Huffman decoder for entropy decoding the corresponding data of the encoded data output from the means, and inverse quantum for inverse quantizing the output from the Huffman decoder , An inverse discrete cosine transformer for performing an inverse discrete cosine transform on the output of the inverse quantizer, and an image enlarger for enlarging the reduced image data output from the encoding means to the size of the original image data. And an adder for adding the image data magnified by the image magnifier and the image data generated by the inverse discrete cosine transform to generate decoded image data in the decoding means. Disclosed is an image encoding device.

[0009]

BEST MODE FOR CARRYING OUT THE INVENTION Embodiments of the present invention will be described below with reference to examples. FIG. 1 is a block diagram showing an example of an embodiment of an image coding apparatus according to the present invention, which is different from the conventional configuration of FIG. 4 in that an image reduction unit 101, an image enlargement unit 102, a difference unit 103, and a decimation unit 104. , A thinning-out decision unit 105, an encoded data insertion unit 106, an adder 107, and an image enlargement unit 108 are added. However, circuits having the same functions as those in FIG. 4 are denoted by the same reference numerals.

2 and 3 are flowcharts showing the operation of the image coding apparatus of FIG. 1, and the operation will be described below with reference to this flowchart. First, in FIG. 2 showing encoding, digitized image data is taken in (step 201), and image data reduction processing is performed by the image reduction device 101 (step 202). In this reduction processing, first, the input image data is divided into, for example, N × N pixel blocks, and each block is subjected to a discrete cosine transform to generate a DC signal.
Obtain the T coefficient X. Next, the DCT coefficient Y consisting of P × P (P <N) on the low frequency side is taken out and subjected to inverse discrete cosine transform, and each pixel value obtained as a result is multiplied by P / N and multiplied by P / N. The reduced image data reduced to is generated. Normally, the above processing is repeated several times to generate reduced image data having a target reduction ratio.

Next, the reduced image data generated as described above is enlarged by the image enlarger 102 to generate image data having the same size as the input image data (step 203). That is, blocks of N × N pixels are sequentially taken out from the reduced image data, a DCT coefficient X is obtained by performing a discrete cosine transform for each of them, and a coefficient having a value “0” is added to the high frequency side to obtain P × P DC (P> N) DC
Let T coefficient Y. Then, the DCT coefficient Y is subjected to inverse discrete cosine transform, and each pixel value obtained as a result is multiplied by P / N to generate magnified image data magnified P / N times. Normally, even in the case of enlargement, the above processing is repeated several times to generate enlarged image data having a target enlargement ratio. FIG. 6 shows the reduction / enlargement process described above. N = 8, P = 5 for reduction, and N = for enlargement.
8 and P = 10.

The differencer 103 generates difference image data between the input image data and the enlarged image data generated by the image enlarger 102 (step 204). FIG. 7 is an upper body image of a person schematically showing an original image for a test, where hairs, sweaters and the like are represented by black dots, which are actually almost black-painted portions. . When this original image is reduced and then enlarged as described above, information representing the contour is lost in the process of the processing, and the contour of the image after the enlargement processing becomes blurred. Therefore, when the difference from the original image is taken in step 204, this time the image is an image in which the contour part is emphasized, which is schematically shown in FIG.

This difference image data is divided into blocks, and the processing of steps 206 to 212 is repeated for each block (steps 205, 213 and 21).
4). That is, first, like the conventional JPEG, the image data of each block is discrete cosine transformed by the discrete cosine transformer 41 (step 206), and the generated D is generated.
The CT coefficient is quantized by the quantizer 42 (step 207).

Most of the quantized DCT coefficients obtained for the difference image data obtained as described above are distributed near 0, and many blocks are all 0 for one block. Appears. This is the same for both monochrome and color images, and when the quantized DCT coefficients are all 0s, when the blocks are converted into pixel values, the pixel values are all 0s. Actually, when the differential image as shown in FIG. 8 is compressed and expanded by the conventional JPEG method, as shown in FIG. 9, the part where the pixel value is 0 (the hatched part in FIG. 9).
Appears widely. Therefore, the decimator 104 checks whether all the quantized DCT coefficients are 0 and stores the result in the decimating table (steps 208 to 21).
0). The thinning-out table is a table for storing a flag for each block. The flag = “0” means that the quantized DCT coefficients of the block are all 0, and the flag = “1” does not. It means that. The Huffman encoder 43 performs entropy coding on the block of the flag = “1” in the thinning table (step 211) and outputs it as coded data to the communication channel or the storage medium 44 (step). 212). Further, the thinning table and the reduced image data are also output to the communication channel or the storage medium 44. Although the case of performing the normal discrete cosine transform has been described above, the above thinning is also effective in the case of the adaptive discrete cosine transform that encodes the difference value of the DC components of the adjacent blocks.

In the decoding process shown in the flow chart of FIG. 3, first, the encoded data and reduced image data and the thinning table are fetched from the communication channel or the storage medium 44 (steps 301 and 302), and for each block. The processes of steps 304 to 312 are repeated (steps 303, 313, 314). That is, first, the thinning-out decision device 10
5, the thinning table is checked, and in the case of the block whose flag is “0”, the encoded data inserter 106
To insert the encoded data (step 30)
4). When the coded data inserter 106 receives this instruction, it sends the coded data in which all DCT coefficients are 0 to the Huffman decoding 45 (step 305). The Huffman decoder 45 receives the DCT from the encoded data inserter 106.
The encoded data of the coefficient or its corresponding flag or the block of "1" fetched from the communication channel or the storage medium 44 is entropy-decoded (step 306). The decoded data is inversely quantized by the inverse quantizer 46 (step 307), and then the inverse discrete cosine transformer 47 is used.
Is subjected to inverse discrete cosine transform (step 308) and output as difference image data (step 309).

Next, the reduced image data from the communication channel or the storage medium 44 is enlarged by the image enlarger 108 in the same manner as the image enlarger 102 (step 310), and the adder 10 is added.
In step 7, the enlarged image data and the decoded difference image data output in step 309 are added to generate decoded image data (step 311), which is output (step 312). In this way, the reduced image data generated according to the flowchart of FIG. 2 can be decompressed / decoded by performing the decoding process while scanning the thinning table for each block. From the processing result of the original image shown in FIG. 7, it was confirmed that the encoding / decoding was performed without significant deterioration of the image quality.

Next, the compression rate in the above-described embodiment will be examined. 640 x 400 pixels for one still image,
If there are 24 planes, the data amount is 6144 Kb (kilobit). When this is compressed to the maximum possible compression rate of about 1/50 of that of the conventional JPEG, it is about 123 Kb.
On the other hand, if the same image data is divided into blocks of 8 × 8 pixels and processed, then 640 × 400/64 = 400.
Since it is 0 block, if 1 bit as a flag is given to each block, the thinning table is 4 Kb.
Will be the size of. The reduced image data is 6144, assuming that the input image data is compressed to 1/16 both vertically and horizontally.
kb / (16 × 16) = 24 kb. The other output data, the encoded data, is generated only for blocks having non-zero DCT coefficients after quantization, so that it is 4000x blocks (0≤x≤1), and each block is different from the conventional one. If the same JPEG is used and compressed to 1/50, the total amount of encoded data is 6144x / 50≈12.
It becomes 3xkb. From the above, the total data amount Z sent to the communication path or the storage medium according to the present embodiment is

## EQU1 ## Z≈123x + 24 + 4 = 123x + 28 kb. Therefore, when some specific values of Z for the value of x are calculated and the ratio R of the 1/50 compression result of only conventional JPEG to 123 kb is calculated, approximate values are as shown in Table 1.

[Table 1] As is clear from Table 1, in the present example, the block in which the DCT coefficients of the output of the quantizer 42 are all 0 is 23% (x = 0.7.
7) In some cases, the compression rate is almost the same as JPEG, but 5
When 0% (x = 0.5) 73 when JPEG only
%, When it is 75% (x = 0.25), it is 48%, that is, about half that of JPEG only. Moreover, in an actual example, x <0.5 is common, and according to the present embodiment, it is possible to perform a much wider compression than in the case of JPEG.

In the above embodiment, the JPE
Similar to G, the differential image data is converted by the discrete cosine transform and Huffman coding is performed as the entropy coding method, but the same effect can be obtained by other methods.

[0019]

According to the present invention, still image data of a natural image can be compressed more than ever before without deteriorating the image quality.

[Brief description of drawings]

FIG. 1 is a block diagram showing an embodiment of an encoding device according to the present invention.

2 is a flowchart showing a flow of an encoding process in the apparatus of FIG.

3 is a flowchart showing a flow of a decoding process in the device of FIG.

FIG. 4 is a block diagram showing a conventional JPEG encoding apparatus.

FIG. 5 is a flowchart showing a flow of encoding / decoding processing in the conventional JPEG system.

FIG. 6 is an explanatory diagram of a method of reducing and enlarging image data.

FIG. 7 is a diagram schematically showing an example of an original image to be processed.

FIG. 8 is a diagram schematically showing a difference image between the original image of FIG. 7 and its enlarged image.

9 is a diagram schematically showing an image in which the difference image of FIG. 8 is compressed by JPEG and then expanded.

[Explanation of symbols]

 41 Discrete Cosine Transform 42 Quantization 43 Huffman Encoder 45 Huffman Decoder 46 Inverse Quantizer 47 Inverse Discrete Cosine Transform 101 Image Reducer 102 Image Enlarger 103 Differencer 104 Decimator 105 Decimator 105 Encoding Data inserter 107 Adder 108 Image enlarger

Claims (8)

[Claims] 1. The reduced image data is generated from the input original image data, the reduced image data is enlarged to the same size as the original image data to generate enlarged image data, and the enlarged image data and the Difference image data representing a difference from the original image data is generated, the difference image data is divided into a plurality of blocks, orthogonal transformation and quantization are performed on each block, and all the quantized data are 0. A table is generated in which a flag having "0" for each block that has become "0" and a value that has "1" for all blocks that have not become 0 is provided for each block, and the flag is " A code that encodes the quantized data only for the block that becomes 1 "to generate the encoded data, and thus outputs the reduced image data, the encoded data, and the table. While performing the encoding process, the reduced image data output by the encoding process,
The coded data and the table are input, the coded data corresponding to the block is targeted for the block of which the flag is "1", and the pixel value of the block of which the flag is "0" is Decodes the coded data when it is constant, dequantizes the decoded data, and inverse-orthogonally transforms the output to generate decoded differential image data corresponding to the differential image data. Then, the input reduced image data is enlarged to the same size as the original image, and thereafter, the pixel value of each of the decoded differential image data is added to perform a decoding process of generating reproduced image data of the original image. Characteristic image coding method. 2. The image coding method according to claim 1, wherein the orthogonal transform is a discrete cosine transform, and the inverse orthogonal transform is an inverse discrete cosine transform. 3. The orthogonal transform is an adaptive discrete cosine transform that encodes a difference between DC components of adjacent blocks, and the inverse orthogonal transform is an inverse adaptive discrete cosine transform. Image coding method. 4. The image coding method according to claim 1, wherein the coding is entropy coding, and the decoding is entropy decoding. 5. The image coding method according to claim 4, wherein the entropy coding and the entropy decoding are Huffman coding and Huffman decoding. 6. Enlargement of image data in the encoding process and the decoding process is performed by dividing the image data into a plurality of blocks and subjecting each block to a discrete cosine transform,
It is characterized in that the coefficient of the high frequency part of the generated block of DCT coefficients is added, the block obtained by the addition is converted into a block of pixel values by inverse discrete cosine transform, and brightness adjustment is performed. The image coding method according to claim 1. 7. The reduction of the image data in the encoding process is performed by dividing the image data into a plurality of blocks and subjecting each block to a discrete cosine transform to generate a DC signal.
2. The coefficient in the high frequency part of the block of T coefficient is deleted, the block obtained by the deletion is converted into a block of pixel values by inverse discrete cosine transform, and brightness adjustment is performed. The image coding method described in. 8. An image reducer for generating reduced image data from input original image data, and an image enlarger for expanding the reduced image data to the size of the original image data to generate an enlarged image. A difference unit for generating difference image data by taking a difference between the enlarged image generated by the image enlarger and original image data; and a DCT coefficient obtained by discrete cosine transforming the difference image data in block units. A discrete cosine transformer for generating, a quantizer for quantizing the DCT coefficients, a block for which the quantized DCT coefficients are all 0, "0", and a block for which all are not 0 On the other hand, "1"
And a Huffman encoding for entropy-encoding the quantized DCT coefficient of the block whose flag is set to "1". Determination unit and an inserter for generating insertion coded data such that the pixel value of the block having the flag "0" is constant, and the flag For the block of which "0" is the insertion coded data from the thinning decision and the inserter, for the block of which the flag is "1", the corresponding data of the coded data output from the coding means Huffman decoder for entropy decoding, dequantizer for dequantizing the output from the Huffman decoder, and inverse discrete for the output of the dequantizer. Inverse discrete cosine converter for sine conversion, image enlarger for enlarging reduced image data output from the encoding means to the size of original image data, and image data enlarged by the image enlarger And an image data generated by performing the inverse discrete cosine transform to generate decoded image data, and an adder for decoding the image data.