JP4144749B2 - Image compression / decompression system, image compression apparatus, image expansion apparatus, image compression method, and image expansion method - Google Patents

Description

The present invention relates to an image compression / decompression system, an image compression apparatus and an image expansion apparatus thereof, and an image compression method and an image expansion method .

  There are many image compression / decompression processing methods that have been established as international standards or are used as de facto standards, but since they are widely used only after being recognized by many users, both methods impose a burden on users. A mechanism that does not apply is adopted. Specifically, a highly versatile quantization table is empirically found so that almost all images can be compressed and expanded while maintaining a certain level of image quality. It is used as a common quantization table. Normally, this quantization table is built in an image compression apparatus or an image expansion apparatus, or is embedded in a code string output from the image compression apparatus. FIG. 16 shows an example in which a quantization table is embedded between header data and payload data.

  However, fixing the quantization table in this way causes the following problems. One is a problem related to information security, and it is difficult to protect images that are not desired to be disclosed. The other is a problem related to image quality, in which image data cannot be optimally encoded in accordance with image characteristics and user preferences.

  As for the former security problem, a method of scrambling code data and a method of providing an authentication key have been conventionally known, but the processing is complicated and it is difficult to completely prevent intrusion from the outside. .

  The latter problem related to image quality is that only one axis called “compression rate” is seen, and image compression is performed without considering important axes such as “motion amount” and “subband code amount distribution”. For this reason, the phenomenon that a high-definition image region is blurred or a code amount is large despite an image having a small amount of motion often occurs.

  Note that JPEG 2000, which became an international standard in 2001, has attracted attention as a successor algorithm to the still-image compression method JPEG that is now widely used. A technique for controlling the quantization of wavelet coefficients using motion information and texture detail information in units of blocks is described in Patent Document 1 regarding an image compression apparatus using JPEG2000.

  Here, an outline of JPEG2000 will be described in order to understand the description of the embodiment of the present invention. FIG. 10 is a block diagram of a system for realizing the hierarchical encoding algorithm that is the basis of JPEG2000. This system includes functional blocks of a color space conversion / inverse conversion unit, a two-dimensional wavelet transform / inverse conversion unit, a quantization / inverse quantization unit, an entropy encoding / decoding unit, and a tag processing unit. .

  One of the biggest differences between this system and the conventional JPEGk algorithm is the frequency conversion method. JPEG uses discrete cosine transform (DCT) for frequency conversion. On the other hand, in the hierarchical encoding algorithm of JPEG2000, discrete wavelet transform (DWT) is used for frequency conversion. DWT has the advantage that the image quality in the high compression region is better than DCT, and this is one of the main reasons why DWT is adopted in JPEG2000.

  Another major difference is that in the JPEG 2000 hierarchical encoding algorithm, a functional block called a tag processing unit for performing code formation is added to the final stage of the system. In this tag processing unit, compressed data is generated as code string data during an image compression operation, and code string data necessary for decompression is interpreted during the decompression operation. JPEG 2000 can implement various convenient functions based on the code string data. For example, the compression / decompression operation can be freely stopped at an arbitrary hierarchy (decomposition level) corresponding to octave division of DWT on a block basis (see FIG. 12).

In JPEG2000, as shown in FIG. 10, a color space conversion unit / inverse conversion unit is often connected to an input / output portion of an original image. For example, RGB color system consisting of R (red) / G (green) / B (blue) components of primary colors and Y (yellow) / M (magenta) / C (cyan) components of complementary colors From YMC color system consisting of YUV
Or the part which performs conversion to a YCbCr color system or reverse conversion corresponds to this.

  Next, the flow of compression (encoding) processing in JPEG2000 will be described. In general, as shown in FIG. 11, in a color image, each component (in this case, the RGB primary color system) of the original image is divided by a rectangular area. This divided rectangular area is called a tile. In the example of FIG. 11, each component is divided into a total of 16 rectangular tiles, 4 × 4 in length and width. Such individual tiles serve as a basic unit for executing the image data compression / decompression process, and the image data compression / decompression operation is performed independently for each component and for each tile.

  When encoding image data, the data of each tile of each component is subjected to color space conversion by the color space conversion / inverse conversion unit, and then two-dimensional wavelet conversion (in order) by the two-dimensional wavelet conversion / inverse conversion unit. (Transformation) is applied and space-divided into frequency bands (subbands).

FIG. 12 shows subbands at each decomposition level when the number of decomposition levels is three. That is, the tile original image (0LL) (decomposition level 0) obtained by tile division of the original image is subjected to two-dimensional wavelet transform, and the decomposition level 1 subband (1LL,
1HL, 1LH, 1HH) are separated (b). Subsequently, the low-frequency component 1LL in this hierarchy is subjected to two-dimensional wavelet transform, and the decomposition level 2 subband (2LL, 2LL,
2HL, 2LH, 2HH). Similarly, the two-dimensional wavelet transform is applied to the low frequency component 2LL, and the decomposition level 3 subband (3LL,
3HL, 3LH, 3HH). A large composition level corresponds to an upper hierarchical level, and a small decomposition level corresponds to a lower hierarchical level.

  Next, the bits to be encoded are determined in the designated encoding order, and the context is generated from the bits around the target bits by the quantization / inverse quantization unit.

  The wavelet coefficients quantized by the quantization / inverse quantization unit are divided into non-overlapping rectangles called “precincts” for each subband. This was introduced to use memory efficiently in implementation. As shown in FIG. 15, one precinct is composed of three rectangular regions that are spatially matched. Each precinct is divided into non-overlapping rectangular “code blocks”. This is a basic unit when entropy coding is performed.

  The entropy encoding / decoding unit performs encoding on the tile of each component by probability estimation from the context and the target bit. In this way, encoding processing is performed in tile units for all components of the original image. Finally, the tag processing unit combines all the encoded data from the entropy encoding / decoding unit into one code string data, and adds a tag and tag information to the code string data, thereby performing FIG. Generate a codestream with the structure shown. As shown in the figure, tag information called a header is added to the head of the code stream and the head of the code data (bit stream) of each tile, followed by the encoded data of each tile. A tag (end of codestream) is placed again at the end of the code string data.

  FIG. 14 shows a code stream structure when a packet containing encoded wavelet coefficient values is represented for each subband. (A) is a structure when tile division is not performed, and (b) is a structure when divided into four tiles. Regardless of the presence or absence of tile division, the packet sequence structure is similar to that illustrated in the lowermost part of FIG.

  On the other hand, at the time of decoding (decompression), contrary to the time of encoding, image data is generated from the code string data of each tile of each component. In this case, the tag processing unit interprets tag information added to the code string data input from the outside, decomposes the code string data into code string data of each tile of each component, and codes the tiles of each component. Decoding processing is performed for each column data. At this time, the positions of the bits to be decoded are determined in the order based on the tag information in the code string data, and the quantization / inverse quantization unit performs the peripheral bits of the target bit position (decoding has already been completed). Context) is generated from the list of In the entropy encoding / decoding unit, decoding is performed by probability estimation from the context and code string data to generate a target bit, and the target bit is written in the position of the target bit. Since the data decoded in this way is spatially divided for each subband, the two-dimensional wavelet inverse transform / inverse transform unit performs the two-dimensional wavelet inverse transform on each component of the image data. Each tile is restored. The restored data is converted into original color system data by the color space conversion / inverse conversion unit.

  The above is the outline of the algorithm of JPEG2000. The Motion-JPEG2000 algorithm is an extension of the JPEG2000 algorithm for a still image, that is, a single frame to a plurality of frames.

JP 2001-326936 A

  The object of the present invention is to provide easy and reliable control of non-disclosure / disclosure of compressed images without departing from the international standard image compression / decompression method and the image compression / decompression method used as a de facto standard. It is to enable compression of optimal image data according to the characteristics of the image and the user's preference.

According to the first aspect of the present invention, there is provided an image compression device that generates a code string by performing compression processing of image data, an image expansion device that generates image data by performing expansion processing of a code string generated by the image compression device, In an image compression / decompression system in which a quantization table group storage device storing a quantization table group is connected to a network,
The image compression apparatus obtains a quantization table corresponding to one or more feature amounts related to image data and a specified compression rate from the quantization table group storage apparatus, and the quantization table is included in a compression process. Generating a code string having header data in which information for identifying the used quantization table is described, which is used for processing, and does not include the quantization table;
The image expansion device acquires a quantization table from the quantization table group storage device based on information described in header data in a code string generated by the image compression device, and stores the quantization table. An image compression / decompression system used for inverse quantization processing included in decompression processing.

The invention of claim 2 is an image compression apparatus that can be used in the image compression / decompression system according to the invention of claim 1,
Image input means for inputting image data;
Compression processing means for generating a code string by performing compression processing at a specified compression rate on the image data input by the image input means;
Analysis means for obtaining one or more feature quantities related to the image data input by the image input means;
A quantization table acquisition unit that acquires a quantization table corresponding to the one or more feature amounts obtained by the analysis unit and the specified compression rate from a quantization table group storage device through a network;
Code string output means for outputting the code string generated by the compression processing means;
Have
The compression processing means uses the quantization table acquired by the quantization table acquisition means for quantization processing included in the compression processing, and uses header data in which information for identifying the quantization table is described. An image compression apparatus characterized by generating a code string having no quantization table.

The invention of claim 3 is an image compression apparatus usable in the image compression / decompression system according to the invention of claim 1,
Image input means for inputting image data;
Analysis means for obtaining one or more feature quantities related to the image data input by the image input means;
Frequency conversion means for generating a frequency conversion coefficient hierarchically divided into frequency bands by frequency-converting the image data input by the image input means for each of one or more rectangular regions;
A quantization table acquisition means for acquiring a quantization table corresponding to the one or more feature amounts obtained by the analysis means and a specified compression ratio from a quantization table group storage device through a network;
Quantization means for quantizing the frequency conversion coefficient generated by the frequency conversion means using the quantization table acquired by the quantization table acquisition means;
Encoding means for encoding the frequency coefficient after quantization by the quantization means;
The encoded data by the encoding means is used as payload data, and a code string formed by combining header data in which information for identifying a quantization table used for quantization is described at the head is created. Code string creating means;
Code string output means for outputting the code string created by the code string creating means;
It is an image compression apparatus characterized by having.

  The invention according to claim 4 is characterized in that the analysis means is means for obtaining a motion amount and a frequency band code amount distribution as a feature amount from the frequency conversion coefficient generated by the frequency conversion means. Is an image compression apparatus.

  In the invention of claim 5, the quantization table has a quantization index value corresponding to each value of the motion amount, frequency band code amount distribution, and compression rate, and the header data in the code string includes the motion amount, the frequency band. 5. The image compression apparatus according to claim 4, wherein quantization index values corresponding to respective values of code amount distribution and compression rate are described.

  The invention according to claim 6 is the image compression apparatus according to claim 5, wherein the quantization index value corresponding to the motion amount is obtained from the correlation coefficient value of the frequency conversion coefficient between successive frames. .

  The invention of claim 7 is characterized in that a quantization index value corresponding to a frequency band code amount distribution is obtained from a histogram representing a code amount sum of frequency conversion coefficients included in each frequency band or a plurality of frequency bands. An image compression apparatus according to the invention of claim 5.

  An eighth aspect of the present invention is the image compression apparatus according to the third aspect of the present invention, wherein the frequency transform is a discrete wavelet transform.

The invention of claim 9 is an image decompression apparatus usable in the image compression / decompression system according to the invention of claim 1,
Code string input means for inputting a code string which is compressed data of image data;
Decompression processing means for generating image data by decompressing the code string input by the code string input means;
Image output means for outputting the image data generated by the decompression processing means;
A quantization table for obtaining a quantization table from a quantization table group storage device through a network according to information for identifying the quantization table described in header data in the code string input by the code string input means Acquisition means;
Have
The decompression processing means is an image decompression apparatus using the quantization table acquired by the quantization table acquisition means for the inverse quantization process included in the extension processing.

The invention of claim 10 is an image decompression apparatus that can be used in the image compression / decompression system according to the invention of claim 1,
Code string input means for inputting a code string which is compressed data of image data;
Syntax analysis means for separating header data and payload data in the code string input by the code string input means, and analyzing information described in the header data;
Decoding means for decoding payload data in the code string based on the analysis result of the syntax analysis means;
Quantization table acquisition means for acquiring a quantization table from a quantization table group storage device through a network according to information for identifying the quantization table extracted from the header data by the syntax analysis means;
Inverse quantization means for inversely quantizing the frequency transform coefficient decoded by the decoding means using the quantization table acquired by the quantization table acquisition means;
Frequency inverse transform means for performing frequency inverse transform on the frequency transform coefficient inversely quantized by the inverse quantization means;
And an image output means for outputting image data generated by the frequency inverse transform of the frequency inverse transform means.

The invention of claim 11 is an image input step for inputting image data;
A compression processing step for generating a code string by performing compression processing at a specified compression rate on the image data input in the image input step;
An analysis step for obtaining one or more feature quantities related to the image data input by the image input step;
A quantization table acquisition step of acquiring a quantization table corresponding to the one or more feature amounts obtained by the analysis step and the specified compression rate from a quantization table group storage device through a network;
A code string output step for outputting the code string generated by the compression processing step;
Have
The compression processing step uses the quantization table acquired by the quantization table acquisition step for quantization processing included in the compression processing, and converts header data in which information for identifying the quantization table is described. This is an image compression method characterized by generating a code string that does not include a quantization table.

The invention of claim 12 includes an image input step of inputting image data;
An analysis step for obtaining one or more feature quantities related to the image data input by the image input step;
A frequency conversion step of generating a frequency conversion coefficient that is divided into frequency bands hierarchically by frequency-converting the image data input by the image input step for each of one or more rectangular regions;
A quantization table acquisition step of acquiring a quantization table corresponding to the one or more feature amounts obtained by the analysis step and the specified compression ratio from a quantization table group storage device through a network;
A quantization step for quantizing the frequency conversion coefficient generated by the frequency conversion step using the quantization table acquired by the quantization table acquisition step;
An encoding step of encoding the frequency coefficient after quantization by the quantization step;
A code string formed by combining encoded data obtained in the encoding step with payload data and header data in which information for identifying a quantization table used for quantization is described at the head is created. A code string creation step;
A code string output step for outputting the code string created in the code string creation step;
An image compression method characterized by comprising:

  According to a thirteenth aspect of the present invention, in the image compression according to the twelfth aspect, the analysis step obtains a motion amount and a frequency band code amount distribution as a feature amount from the frequency conversion coefficient generated by the frequency conversion step. Is the method.

  In the invention of claim 14, the quantization table has a quantization index value corresponding to each value of motion amount, frequency band code amount distribution, and compression rate, and the header data in the code string includes the motion amount, frequency band. 14. The image compression method according to claim 13, wherein a quantization index value corresponding to each value of code amount distribution and compression rate is described.

The invention of claim 15 is a code string input step of inputting a code string which is compressed data of image data,
An expansion processing step for generating image data by performing an expansion process on the code string input in the code string input step;
An image output step for outputting the image data generated by the decompression step;
A quantization table for obtaining a quantization table from a quantization table group storage device through a network according to information for identifying the quantization table described in header data in the code string input by the code string input step An acquisition step;
Have
In the image expansion method, the expansion processing step uses the quantization table acquired in the quantization table acquisition step for the inverse quantization processing included in the expansion processing.

The invention of claim 16 includes a code string input step of inputting a code string that is compressed data of image data;
A syntax analysis step for separating header data and payload data in the code string input by the code string input step, and analyzing information described in the header data;
A decoding step of decoding the payload data in the code string based on the analysis result of the parsing step;
A quantization table obtaining step for obtaining a quantization table from a quantization table group storage device through a network according to information for identifying a quantization table extracted from the header data by the parsing step;
An inverse quantization step of inversely quantizing the frequency transform coefficient decoded by the decoding step using the quantization table acquired by the quantization table acquisition step;
A frequency inverse transform step for performing frequency inverse transform on the frequency transform coefficient inversely quantized by the inverse quantization step;
And an image output step of outputting image data generated by the frequency inverse transform of the frequency inverse transform step.

  According to the present invention, it is possible to easily and reliably manage non-disclosure / disclosure of a compressed image without departing from an international standard method or a de facto standard method for image compression / decompression. In addition, since there is no need to fix the quantization table used for compression, the optimal quantization according to the characteristics of each image is considered, taking into account the feature amount such as the amount of motion and frequency band code amount distribution for each region of the image. A table can be used, and compression with good image quality can be achieved.

  FIG. 1 shows an example of an image compression / decompression system of the present invention. Reference numeral 100 denotes an image compression apparatus of the present invention that performs compression processing of image data and outputs a code string, 200 denotes an image expansion apparatus of the present invention that expands a code string obtained by compressing image data into image data, and 300 denotes an image compression apparatus 100. In addition, a quantization table group storage device 302 that stores a quantization table group that can be used by the image expansion device 200 is a code string storage device 302 for storing a compressed code string. These devices are connected to a network 303 (for example, a LAN, an intranet, the Internet, etc.) and can communicate with each other via the network 303. Although one image compression apparatus 100 and one image expansion apparatus 200 are shown, the number is arbitrary. The image decompression apparatus 200 can fetch the compressed code string directly by the image compression apparatus 100 or can fetch the code string from the code string storage device 302.

  The image compression apparatus 100 includes an image input unit that inputs image data, a compression processing unit that performs compression processing of a specified compression rate on the input image data to generate a code string, and one or more feature quantities related to the input image data And a quantization table acquisition unit that acquires, from the quantization table group storage device 300, the quantization table corresponding to the one or more feature amounts obtained by the analysis unit and the specified compression rate, through the network 303. And a code string output means for outputting the code string generated by the compression processing means to the code string storage device 302 or the image expansion device 200 via the network 303, and the compression processing means is included in the compression processing. Use the quantization table acquired by the quantization table acquisition means for the quantization process, and identify the quantization table Because information has the header data described is a configuration for generating a code string which does not contain a quantization table. A more specific configuration will be described later.

  The image decompression apparatus 200 includes a code string input unit that inputs a code string from the code string storage device 302 or the image compression apparatus 100 through the network 303, and a decompression processing unit that decompresses the input code string and generates image data. According to the information for identifying the quantization table described in the header data in the code string input by the code string input means and the image output means for outputting the image data generated by the decompression processing means , A quantization table acquisition unit that acquires a quantization table from the quantization table group storage device 300 via the network 303, and the expansion processing unit performs the quantization table in the inverse quantization process included in the expansion process. In this configuration, the quantization table acquired by the acquisition unit is used. A more specific configuration will be described later.

  In the code string output from the image compression apparatus 100 of the present invention, the quantization table used for the quantization process at the time of compression is not embedded. The image decompression device that can normally decompress the code string is limited to the image decompression device 200 of the present invention that can acquire the quantization table from the quantization table group storage device 300. An image decompression apparatus that cannot access the quantization table group storage apparatus 300, or even if the access is possible, the quantization table child identification information described in the header data in the code string is correctly interpreted and the quantization table concerned. An image decompression device that does not have the function of acquiring the code cannot normally decompress the code string. In order to make the compressed image undisclosed, the quantization table group storage device 300 on the network itself may be hidden or access to the corresponding quantization table may be restricted. At this time, the code string is modified. There is no need. In this way, non-disclosure and disclosure of compressed images can be managed easily and reliably. In addition, since it is not necessary to fix the quantization table used for compression, it is possible to perform optimum quantization in consideration of the characteristics of individual images.

  Embodiments of an image compression apparatus and an image expansion apparatus according to the present invention will be described below. Here, a description will be given assuming that a JPEG2000 algorithm is used as the compression / decompression algorithm. It is assumed that the image to be handled is a Motin still image, and the still image of each frame is independently compressed and decompressed by the JPEG2000 algorithm.

  FIG. 2 is a block diagram showing an embodiment of the image compression apparatus 100 of the present invention, and FIG. 3 is a block diagram showing a specific configuration example of the image analysis means 107 in FIG.

  In FIG. 2, 101 is an image input means for inputting image data. The frequency transform unit 102 divides input image data into one or more tiles (rectangular regions), performs two-dimensional discrete wavelet transform for each tile, and hierarchically divides the wavelet transform coefficients into subbands (frequency bands). (Frequency conversion coefficient) is generated. The quantization unit 103 quantizes the coefficient of each subband using the quantization table acquired by the quantization table acquisition unit 108. The quantized wavelet transform coefficients are entropy-coded in bit plane units for each subband by the encoding means 104. The code string creation unit 105 uses the code output from the coding unit 104 as payload data, and header data in which information for identifying the quantization table used by the quantization unit 103 is described at the beginning. A code string in which the quantization table is not embedded is created. This code string is output by the code string output means 106.

  The image analysis unit 107 is a unit that obtains one or more feature amounts related to input image data for each tile. In this embodiment, the image analysis means 107 obtains a motion amount and a subband (frequency band) code amount distribution as a feature amount, and has a configuration as shown in FIG. In FIG. 3, the packet length detection means 120 detects the length of the packet obtained by the frequency conversion means 102 for each tile. The packet length storage unit 122 is a unit that temporarily stores the detected packet length. The difference detection unit 123 detects the difference between the packet length of the current frame and the packet length of the previous frame stored in the packet length storage unit 122 for each tile. This difference in packet length is a correlation coefficient value of wavelet transform coefficients between successive frames, and this is input to the quantization table acquisition means 108 as a quantization index value (motion amount index value) corresponding to the motion amount. The The subband code amount distribution detecting unit 121 obtains a histogram of the sum of packet lengths (sum of code amounts of wavelet transform coefficients) for each subband or for each subband, and a quantization index value corresponding to the subhand code amount distribution. (Subband code amount distribution index value) is output. The quantization table acquisition unit 108 is a quantization table group storage device that stores a quantization index value (compression rate index value), a motion amount index value, and a subband code amount index value corresponding to the compression rate designated by the user via the network 303. 300, and receives one or more quantization tables having a corresponding quantization index value from the quantization table group storage device 300. The received quantization table is given to the quantizing means 103 and used for the wavelet transform coefficient quantization processing as described above.

  In this embodiment, the quantization table identification information described in the header data in the code string created by the code string creation unit 105 is used as the quantization table quantization index value (compression ratio index value). , Motion amount index value, subhand code amount distribution index value).

  A schematic diagram of a quantization table group stored in the quantization table group storage device 300 is shown in FIG. As shown in the figure, the quantization table includes quantization index values (compression rate index value, motion amount index value, subband code amount distribution index value) corresponding to each value of motion amount, sub baton code amount distribution, and compression rate. Have. The compression rate index value includes a case of lossless compression. Conventionally, as shown in FIG. 6, quantization is performed using a quantization table that considers only one axis of compression rate. On the other hand, by using a quantization table in consideration of the three axes of motion amount, subband code amount distribution, and compression rate, it is possible to perform image compression with good image quality. This will be described later. To do.

  In the above, the quantization index values of the motion amount, subband, and compression rate for each tile are transmitted to the quantization table group storage device 300, and one or more quantization tables having these quantization index values are acquired. Then explained. However, by transmitting, for example, only the compression rate index value from the quantization table acquisition unit 108, a quantization table group having the same compression rate index value (for example, the quantization table group shown in the lower diagram of FIG. 5) is acquired. Of course, a quantization table having a motion amount index value and a subband code amount index value may be selected for each tile from the quantization table group and given to the quantization means 103. Such an embodiment is also included in the present invention. This is the same in the case of the image expansion apparatus described later.

  FIG. 4 is a block diagram showing an embodiment of the image expansion apparatus 200 according to the present invention. In FIG. 4, reference numeral 201 denotes a code string input means for inputting a code string that is compressed data of image data. The syntax analysis unit 202 separates the header data and the payload data in the input code string, and analyzes the information described in the header data. Based on the analysis result of the syntax analysis unit 202, the decoding unit 203 decodes the payload data in the code string and generates a wavelet transform coefficient (frequency transform coefficient). The inverse quantization unit 204 performs inverse quantization of the wavelet transform coefficient decoded by the decoding unit 203 using the quantization table acquired by the quantization table selection unit 207. The frequency inverse transform unit 205 applies the two-dimensional discrete wavelet inverse transform (frequency transform coefficient) to the wavelet transform coefficient inversely quantized by the inverse quantization unit 204, and returns the image data. This image data is output by the image output means 206.

  The parsing unit 202 extracts information (in this case, index values of motion amount, subband code amount distribution, and compression rate) for identifying the quantization table from the header data, and obtains the quantization table. It gives to the means 207. The quantization table acquisition unit 207 is the same unit as the quantization table acquisition unit 108 (FIG. 2) of the image compression device, and acquires one or more corresponding quantization tables from the quantization table group storage device 300 via the network. , It is given to the inverse quantization means 204.

  Here, the merit of using the quantization table as shown in FIG. 5 will be described with reference to FIGS.

  Conventionally, the compression of Motion still images has been focused only on “quantization” optimized for still images in a frame, and “quantization” has not been performed in consideration of the visual characteristics unique to moving images. It was. Originally, the amount of motion between frames differs from one image region to another, but conventionally, it has been assumed that the amount of motion is uniform over the entire image region. As a result, completely new image quality degradation, which could not be expected when displaying a still image, often appeared when displaying a moving image.

  FIG. 8 shows a conventional quantization method in which the amount of motion is ignored and the subband codes are uniformly reduced. Here, the vertical axis represents the amount of motion, and the correlation coefficient value obtained from the difference of the wavelet coefficient values between the current frame and the previous frame is used. The horizontal axis represents the unit of tile. Note that the numbers (tiles #) given to the tiles are those defined in FIG.

  Looking at the amount of movement, as seen in the upper diagram of FIG. 8, the tile A is large and the tile C is small. However, in the conventional method, the entire image is replaced with a uniform amount of motion. In this example, as shown in the lower diagram of FIG. 8, the movement amounts of all tiles are treated as “none”. Therefore, the only quantization table is referenced regardless of the amount of motion.

  In addition, the quantization of the still image in the frame also uses the only “quantization table” uniquely defined by the compression rate or the bit rate. Originally, although the code amount distribution of the subbands should be different for each image area, conventionally, such consideration has not been taken into consideration. As a result, the image quality of the quantized image greatly fluctuated depending on the properties of the original image, such as an image with many high-frequency components, an image with many low-frequency components, and an image widely distributed from high to low frequencies.

  FIG. 9 shows a conventional quantization method in which the distribution of the code amount is ignored and the subband codes are uniformly reduced. The vertical axis represents the relative code amount when the low-frequency subband code amount in the highest layer (maximum number of decompositin levels) is normalized to “1”. The horizontal axis represents the subband.

  In the curve indicating the code amount distribution for each tile, the thin line portion represents the subband from which the code amount is to be deleted, while the thick line portion represents the subband in which the code amount is stored as it is. ing. (a) corresponds to the case where the number of subbands whose code amount is to be reduced is large, and (b) corresponds to the case where the number of subbands is small. Tile numbers A, B, C, and D are defined in FIG.

  The tiles A and D hold the code data up to the high frequency range. The tile B has a relatively large amount of information in the middle region, and the tile C has a large amount of information in the low region. As described above, even when the subband code amount is different for each tile, the conventional method reduces the subband code data uniformly regardless of the code amount distribution. As a result, especially when the compression ratio is high, the image quality of tiles A, B, and D with a relatively large amount of data in the middle to high range is compared with that of tile C in which data is concentrated in the low range. The deterioration was remarkable.

  In order to specifically describe the above contents, the observation results will be described in detail below for the phenomenon that occurs when Motion-JPEG2000 is used as the compression / decompression algorithm.

  FIG. 7 shows three consecutive frames extracted from the moving image content. In this example, a bicycle that runs against a background of a small amount of movement is shown. It is an image that is relatively easy to distinguish between a dail with a large amount of motion and a small tile. Now focus on the four tiles. Tile # = A is located below the center, tile # = B is located on the upper left, tile # = C is located above the center, and tile # = D is located on the upper right.

  When a Motion still image encoded by conventional quantization that does not take into consideration the amount of motion between frames and the code amount distribution of subbands is expanded and displayed, the following phenomenon appears.

Tile # = A:
The bicycle is closest to the shooting point, and the code amount distribution of the high frequency subband is large due to the structural features. Since the code amount of the high frequency sub-band is reduced as the compression rate increases, it becomes difficult to distinguish the spoke portions of the wheel one by one from the still image in the frame. Since the discrimination of the spokes of a moving bicycle is beyond the limit of human visual acuity, the effect on the moving image is not so great.

Tile # = B:
There is almost no movement of the tree located in the middle of the foreground and foreground bicycles. The leaf portion has a large code amount distribution in the middle band. As the compression rate increases, it becomes difficult to distinguish each leaf individually from the still image in the frame. However, the tendency is not as strong as the bicycle spokes in tile # = A.

Tile # = C:
In the distant view, a skyline having a large code amount distribution in the low frequency subband is arranged. Even if the compression rate increases, it is relatively difficult to be affected by the code amount reduction. However, since there is almost no amount of motion, if even a small amount of low-frequency sub-band data is deleted, the blur will spread over the entire still image in the frame. When viewed in a moving image, “smooth distortion” occurs in the background, which is very disturbing.

Tile # = D
There are many contour portions of the tree in the image area. As a result, the code amount distribution differs from tile # = C in that the data amount on the high frequency side is slightly larger. When comparing the image with tile # = C, there is not much difference between the still images in the frame. However, when a moving image is displayed, “smooth distortion” appears more prominently. This seems to be because the amount of the high-frequency component that has been shaved is large.

  According to the present invention, the quantization table is not fixed, and an optimum quantization table as shown in FIG. 5, for example, can be used for quantization according to the characteristics of the image. It is clear from the above description that image quality deterioration can be effectively suppressed.

  The frequency transform used in the present invention is not limited to the discrete wavelet transform, and a discrete cosine transform (DCT) can also be used.

It is a figure which shows the structural example of the image compression / decompression system of this invention. It is a block diagram which shows one Example of the image compression apparatus of this invention. It is a block diagram which shows the structural example of the image analysis means in FIG. It is a block diagram which shows one Example of the image expansion | extension apparatus of this invention. It is explanatory drawing of the quantization table used by this invention. It is explanatory drawing of the quantization table in which only the compression rate was considered. It is explanatory drawing of the image quality degradation relevant to the conventional quantization. It is explanatory drawing regarding the movement amount of some tiles shown in FIG. It is explanatory drawing of the problem by quantizing uniformly with the code amount for every subband about several tiles shown in FIG. , It is explanatory drawing of the system of JPEG2000. It is explanatory drawing of the tile division | segmentation for every component. It is explanatory drawing of the octave division | segmentation by two-dimensional discrete wavelet transform. It is explanatory drawing of the structure of a code stream of JPEG2000. It is explanatory drawing of the code stream structure which represented the packet in which the encoded wavelet coefficient was accommodated for every subband. . It is explanatory drawing of an image, a tile, a precinct, and a code block. It is explanatory drawing of the code sequence with which the quantization table was embedded.

Explanation of symbols

DESCRIPTION OF SYMBOLS 100 Image compression apparatus 101 Image input means 102 Frequency conversion means 103 Quantization means 104 Encoding means 105 Code sequence creation means 107 Image analysis means 108 Quantization table acquisition means 200 Image decompression apparatus 201 Code sequence input means 202 Syntax analysis means 203 Decoding Quantization means 204 inverse quantization means 205 frequency inverse transform means 206 image output means 207 quantization table acquisition means 300 quantization table group storage device 302 code string storage device 303 network

Claims (16)

An image compression apparatus that generates a code string by compressing image data, an image expansion apparatus that generates image data by performing an expansion process of a code string generated by the image compression apparatus, and a quantization table group In an image compression / decompression system in which a quantization table group storage device is connected to a network,
The image compression apparatus obtains a quantization table corresponding to one or more feature amounts related to image data and a specified compression rate from the quantization table group storage apparatus, and the quantization table is included in a compression process. Generating a code string having header data in which information for identifying the used quantization table is described, which is used for processing, and does not include the quantization table;
The image expansion device acquires a quantization table from the quantization table group storage device based on information described in header data in a code string generated by the image compression device, and stores the quantization table. An image compression / decompression system for use in an inverse quantization process included in an expansion process. An image compression apparatus usable in the image compression / decompression system according to claim 1,
Image input means for inputting image data;
Compression processing means for generating a code string by performing compression processing at a specified compression rate on the image data input by the image input means;
Analysis means for obtaining one or more feature quantities related to the image data input by the image input means;
A quantization table acquisition unit that acquires a quantization table corresponding to the one or more feature amounts obtained by the analysis unit and the specified compression rate from a quantization table group storage device through a network;
Code string output means for outputting the code string generated by the compression processing means;
Have
The compression processing means uses the quantization table acquired by the quantization table acquisition means for quantization processing included in the compression processing, and uses header data in which information for identifying the quantization table is described. An image compression apparatus characterized by generating a code string having no quantization table. An image compression apparatus usable in the image compression / decompression system according to claim 1,
Image input means for inputting image data;
Analysis means for obtaining one or more feature quantities related to the image data input by the image input means;
Frequency conversion means for generating a frequency conversion coefficient hierarchically divided into frequency bands by frequency-converting the image data input by the image input means for each of one or more rectangular regions;
A quantization table acquisition means for acquiring a quantization table corresponding to the one or more feature amounts obtained by the analysis means and a specified compression ratio from a quantization table group storage device through a network;
Quantization means for quantizing the frequency conversion coefficient generated by the frequency conversion means using the quantization table acquired by the quantization table acquisition means;
Encoding means for encoding the frequency coefficient after quantization by the quantization means;
A quantization table is formed by combining encoded data by the encoding means as payload data, and header data in which information for identifying the quantization table used for quantization is combined at the head thereof. Code string creating means for creating a code string that is not embedded;
Code string output means for outputting the code string created by the code string creating means;
An image compression apparatus comprising:   4. The image compression apparatus according to claim 3, wherein the analysis unit is a unit that obtains a motion amount and a frequency band code amount distribution as a feature amount from the frequency conversion coefficient generated by the frequency conversion unit.   The quantization table has a quantization index value corresponding to each value of motion amount, frequency band code amount distribution, and compression rate, and the header data in the code string includes the amount of motion, frequency band code amount distribution, and compression rate. The image compression apparatus according to claim 4, wherein a quantization index value corresponding to each value is described.   6. The image compression apparatus according to claim 5, wherein the quantization index value corresponding to the motion amount is obtained from a correlation coefficient value of a frequency conversion coefficient between successive frames.   6. The image according to claim 5, wherein the quantization index value corresponding to the frequency band code amount distribution is obtained from a histogram representing a sum of code amounts of frequency conversion coefficients included in each frequency band or a plurality of frequency bands. Compression device.   The image compression apparatus according to claim 3, wherein the frequency transform is a discrete wavelet transform. An image expansion apparatus usable for the image compression / decompression system according to claim 1,
Code string input means for inputting a code string which is compressed data of image data;
Decompression processing means for generating image data by decompressing the code string input by the code string input means;
Image output means for outputting the image data generated by the decompression processing means;
A quantization table for obtaining a quantization table from a quantization table group storage device through a network according to information for identifying the quantization table described in header data in the code string input by the code string input means Acquisition means;
Have
The image decompression apparatus, wherein the decompression processing unit uses the quantization table acquired by the quantization table acquisition unit for the inverse quantization process included in the decompression process. An image expansion apparatus usable for the image compression / decompression system according to claim 1,
Code string input means for inputting a code string which is compressed data of image data;
Syntax analysis means for separating header data and payload data in the code string input by the code string input means, and analyzing information described in the header data;
Decoding means for decoding payload data in the code string based on the analysis result of the syntax analysis means;
Quantization table acquisition means for acquiring a quantization table from a quantization table group storage device through a network according to information for identifying the quantization table extracted from the header data by the syntax analysis means;
Inverse quantization means for inversely quantizing the frequency transform coefficient decoded by the decoding means using the quantization table acquired by the quantization table acquisition means;
Frequency inverse transform means for performing frequency inverse transform on the frequency transform coefficient inversely quantized by the inverse quantization means;
And an image output means for outputting image data generated by the frequency inverse transform of the frequency inverse transform means. An image input step for inputting image data;
A compression processing step for generating a code string by performing compression processing at a specified compression rate on the image data input in the image input step;
An analysis step for obtaining one or more feature quantities related to the image data input by the image input step;
A quantization table acquisition step of acquiring a quantization table corresponding to the one or more feature amounts obtained by the analysis step and the specified compression rate from a quantization table group storage device through a network;
A code string output step for outputting the code string generated by the compression processing step;
Have
The compression processing step uses the quantization table acquired by the quantization table acquisition step for quantization processing included in the compression processing, and converts header data in which information for identifying the quantization table is described. An image compression method comprising: generating a code string having a quantization table and no quantization table. An image input step for inputting image data;
An analysis step for obtaining one or more feature quantities related to the image data input by the image input step;
A frequency conversion step of generating a frequency conversion coefficient that is divided into frequency bands hierarchically by frequency-converting the image data input by the image input step for each of one or more rectangular regions;
A quantization table acquisition step of acquiring a quantization table corresponding to the one or more feature amounts obtained by the analysis step and the specified compression ratio from a quantization table group storage device through a network;
A quantization step for quantizing the frequency conversion coefficient generated by the frequency conversion step using the quantization table acquired by the quantization table acquisition step;
An encoding step of encoding the frequency coefficient after quantization by the quantization step;
A quantization table formed by combining encoded data obtained by the encoding step as payload data and combining header data in which information for identifying the quantization table used for quantization is described at the head thereof. A code string creation step for creating a code string that is not embedded;
A code string output step for outputting the code string created in the code string creation step;
An image compression method characterized by comprising:   13. The image compression method according to claim 12, wherein the analysis step obtains a motion amount and a frequency band code amount distribution as a feature amount from the frequency conversion coefficient generated by the frequency conversion step.   The quantization table has a quantization index value corresponding to each value of motion amount, frequency band code amount distribution, and compression rate, and the header data in the code string includes the amount of motion, frequency band code amount distribution, and compression rate. The image compression method according to claim 13, wherein a quantization index value corresponding to each value is described. A code string input step for inputting a code string which is compressed data of image data;
An expansion processing step for generating image data by performing an expansion process on the code string input in the code string input step;
An image output step for outputting the image data generated by the decompression step;
A quantization table for obtaining a quantization table from a quantization table group storage device through a network according to information for identifying the quantization table described in header data in the code string input by the code string input step An acquisition step;
Have
The expansion process step uses the quantization table acquired by the quantization table acquisition step for the inverse quantization process included in the expansion process. A code string input step for inputting a code string which is compressed data of image data;
A syntax analysis step for separating header data and payload data in the code string input by the code string input step, and analyzing information described in the header data;
A decoding step of decoding the payload data in the code string based on the analysis result of the parsing step;
A quantization table obtaining step for obtaining a quantization table from a quantization table group storage device through a network according to information for identifying the quantization table extracted from the header data by the parsing step;
An inverse quantization step of inversely quantizing the frequency transform coefficient decoded by the decoding step using the quantization table acquired by the quantization table acquisition step;
A frequency inverse transform step for performing frequency inverse transform on the frequency transform coefficient inversely quantized by the inverse quantization step;
And an image output step of outputting image data generated by the frequency inverse transform of the frequency inverse transform step.

Images