JP2002369202A - Image compression device, image expansion device, image compressing method, image expanding method, program and recording medium stored with the program - Google Patents

Abstract

PROBLEM TO BE SOLVED: To make a tile boundary hardly visible by reducing quantization rate in a region in the vicinity of the tile boundary which is generated, when an image is divided into tiles and compressed images obtained by performing high compression for each tile decoded. SOLUTION: A square tile is designated by an arbitrary size to an image, the image is divided by using the designated square tiles, and the square tiles in which the image is divided are decomposed into bit planes by each tile. The decomposed bit planes are ordered from MSB to LSB according to coding order, and layers, e.g. from the most significant layer 0 to the least significant layer 9 are constituted by using the ordered bit planes. Bit planes of code blocks 2, 3; 6, 7; 10, 11 in a region in the vicinity of the tile boundary, in which region the quantizing rate is suppressed to be relatively low are shifted, so as to be included in a higher rank layer side to bit planes of code blocks 0, 1; 4, 5; 8, 9 in a region distant from the tile boundary in which region the quantizing rate is maintained to be relatively high, and the quantizing rate is reduced.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

The present invention relates to an image compression apparatus, an image expansion apparatus, an image compression method, an image expansion method, a program, and a recording medium on which the program is recorded.
The present invention relates to an image compression device, an image decompression device, an image compression method, an image decompression method, a program that can be executed by a computer, and a recording medium on which the program is recorded, which makes it possible to keep the quantization rate near a tile boundary low.

[0002]

2. Description of the Related Art With the advance of image input technology and output technology thereof, the demand for higher definition of color still images has been extremely increased in recent years. Taking a digital camera (Digital Camera (DC)) as an example of an image input device, the price of a high-performance electric-coupled device (CCD) having more than 3 million pixels has been reduced, and products in the popular price range have been reduced. Has also become widely used. And
A product with 5 million pixels is coming soon. Such high performance of the input device represented by the CCD largely depends on the progress of the silicon process or device technology, and has overcome the trade-off problem of miniaturization and reduction of the S / N ratio. It is said that this increasing trend in the number of pixels will continue for some time.

On the other hand, with respect to image output / display devices, products in the hard copy field such as laser printers, ink jet printers, sublimation printers and the like, and CRTs
In the field of soft copy of flat panel displays such as LCDs, LCDs (liquid crystal display devices), and PDPs (plasma display devices), high definition and low price products are remarkable.

[0004] With the effect of bringing high-performance, low-cost image input / output products to the market, the popularization of high-definition still images has begun. It is expected that the demand for high-definition still images will increase in all scenes in the future. Indeed, the development of technologies relating to networks, including personal computers (PCs) and the Internet, has accelerated these trends. Especially recently, mobile devices such as mobile phones and laptops have become very popular.
Opportunities to transmit or receive high-definition images from any point using communication means are increasing rapidly. Against this background, the demand for higher performance or more functions for image compression / decompression technology that facilitates the handling of high-definition still images will inevitably increase in the future.

At present, JPEG (Joint Photographic) is used as an image compression / decompression algorithm for facilitating the handling of such high-definition still images.
Experts Group) is the most widely used. JPEG20, which became an international standard in 2001,
00 not only has an image compression / decompression algorithm with higher performance than JPEG, but at the same time has a great number of functions and flexibility and expandability for various applications. It is highly expected as a next-generation high-definition still image compression / decompression format succeeding JPEG.

FIG. 17 is a block diagram for explaining the basics of the JPEG algorithm. The JPEG algorithm includes a color space conversion / inverse conversion unit 40, a discrete cosine conversion /
It comprises an inverse transform unit 41, a quantization / inverse quantization unit 42, and an entropy encoding / decoding unit 43. Normally, lossy encoding is used in order to obtain a high compression ratio, and so-called lossless compression and decompression is not performed. Although the original image data is not completely stored, practical problems rarely occur. Therefore, the JPEG method greatly contributes to reducing the memory capacity necessary for compression and decompression processing or storage of image data after compression, and to shorten the time spent for data transmission and reception. Because of these advantages, JPEG has become the most widely used still image compression and decompression algorithm.

FIG. 18 is a block diagram for explaining the basics of the JPEG2000 algorithm. JPEG2
000 algorithm is a color space conversion / inversion unit 50,
It comprises a two-dimensional wavelet transform / inverse transform unit 51, a quantization / inverse quantization unit 52, an entropy encoding / decoding unit 53, and a tag processing unit 54.

As described above, JPEG is the most widely used still image compression / expansion method at present. However, the demand for higher definition still images has not stopped, and the technical limits of the JPEG system are beginning to appear. For example, block noise and mosquito noise that appear on an image that has not been so noticeable until now have become increasingly prominent with the higher definition of the original image. In other words, the deterioration of image quality due to the JPEG file, which has not been a problem in the past, has approached a level that cannot be ignored in practical use. As a result, the improvement of the image quality in the low bit rate, that is, in the high compression rate region has been recognized as the most important problem in the algorithm. JPEG2000
Was born as an algorithm that can solve these problems. And in the near future, the current mainstream JPEG
It is expected to be used together with the format.

One of the biggest differences between FIG. 17 and FIG. 18 is the conversion method. JPEG stands for Discrete Cosine Transform (DCT).
JPEG2000 is the discrete wavelet transform (DWT: Discrete).
Wavelet Transform).
The advantage that DWT has better image quality in a high compression area than DCT is a major reason for adoption in the succeeding algorithm.

Another major difference is that, in the latter case, the tag processing section 5 is used to form a code in the last stage.
4 is added. In this part, compressed data is generated as a code stream during a compression operation, and a code stream necessary for decompression is interpreted during a decompression operation. The codestream has enabled JPEG2000 to realize various convenient functions. For example, FIG. 19 is a diagram showing an example of subbands at each decomposition level when the decomposition level is 3, and in an arbitrary hierarchy corresponding to octave division in the block-based DWT shown in FIG. The compression / decompression operation of the still image can be freely stopped.

In many cases, a color space conversion / inverse conversion unit 40 and a color space conversion / inverse conversion unit 50 are connected to the input / output portion of the original image shown in FIGS. For example, RGB color system composed of R (red) / G (green) / B (blue) components of primary color system, and Y (yellow) / M (magenta) / C (cyan) component of complementary color system Y consisting of
The portion that performs conversion from the MC color system to the YUV or YCrCb color system or vice versa corresponds to this.

Hereinafter, the JPEG2000 algorithm will be described. Here, the definition of terms relating to JPEG2000 is JPEG2000 PartI FDIS.
(Final Draft International Standard). Hereinafter, definitions of typical terms are shown. 1. bit-plane: A two dimensional array of bits.In t
his Recommendation International Standard a bit-pl
ane refers to all the bits of the same magnitude i
n all coefficients or samples.This could refer to
a bit-plane in acomponent, tile-component, code-bloc
k, region of interest, or other. code-block: A rectangular grouping of coefficie
nts from the same subband of a tile-component. decomposition level: A collection of wavelet su
bbands where each coefficient has the same spatial
impact or span with respect to the source compone
nt samples.These include the HL, LH, and HH subbands
of the same twodimensional subband decomposition.
For the last decomposition level theLL subband is
also included. layer: A collection of compressed image data fr
om coding pass of one, or more, code-blocks of a til
e-component.Layers have an order for encoding and
decoding and decoding that must be preserved. precinct: A one rectangular region of a transfo
rmed tile-component, within each resolution level, u
sed for limiting the size of packets.

FIG. 20 is a diagram showing an example of each component of a tiled color image. Color images are
Generally, as shown in FIG. 20, the original components of the image 70, 71, 72 (here RGB primary color system) is a region where the rectangular _(tile) is divided by the _{70 t, 71 t, 72 t} . And individual tiles, eg, R00,
R01, ..., R15 / G00, G01, ..., G15 / B
, B15 are basic units when executing the compression / decompression process. Therefore, the compression / expansion operation is performed independently for each component and for each tile. At the time of encoding, the data of each tile of each component is input to the color space conversion unit 50 shown in FIG. 18 and subjected to color space conversion, and then the two-dimensional wavelet conversion unit 51 performs two-dimensional wavelet conversion (forward conversion). ) Is applied to perform spatial division into frequency bands.

FIG. 19 shows the subbands at each decomposition level when the decomposition level is 3. That is, a two-dimensional wavelet transform is performed on the tile original image (0LL) (decomposition level 0 (60)) obtained by the tile division of the original image, and the decomposition level 1 (61).
(1LL, 1HL, 1LH, 1HH)
Is separated. Subsequently, two-dimensional wavelet transform is performed on the low-frequency component 1LL in this layer to separate the subbands (2LL, 2HL, 2LH, 2HH) indicated by the decomposition level 2 (62). Similarly, the two-dimensional wavelet transform is performed on the low-frequency component 2LL in the same manner, and the decomposition level 3 (6
3) Subbands (3LL, 3HL, 3LH, 3H)
H).

Further, in FIG. 19, the color of the subband to be encoded at each decomposition level is represented by gray. For example, if the decomposition level is 3
, The subbands shown in gray (3HL, 3L
H, 3HH, 2HL, 2LH, 2HH, 1HL, 1L
H, 1HH) is to be encoded, and the 3LL subband is not encoded.

Next, bits to be encoded are determined in the designated order of encoding, and a context is generated from bits around the target bit by the quantization unit 52 shown in FIG. The wavelet coefficient after the quantization process is divided into non-overlapping rectangles called precincts for each subband. This was introduced to make efficient use of memory in the implementation.

FIG. 21 is a diagram for explaining an example of the relationship between precincts and code blocks. In the original image 80, the tile 80 _t0 and the tile 8
It is divided into four tiles: 0 _{t 1} , tile 80 _t2 , and tile 80 _t3 . As shown in FIG. 21, for example, the precinct 80 _p4 is composed of three spatially coincident rectangular regions, and the same applies to the precinct 80 _p6 . Furthermore, each precinct is divided into blocks called non-overlapping rectangular code blocks. In this example, the code block is divided into twelve code blocks from 0 to 11. For example, the code block 80 _b1 indicates the code block number 1. This code block is a basic unit when performing entropy coding.

The coefficient value after the wavelet transform can be quantized and encoded as it is.
In the case of 000, in order to increase the coding efficiency, the coefficient values can be decomposed in units of bit planes, and the bit planes can be ranked for each pixel or code block.

FIG. 22 is a diagram for explaining an example of a procedure for prioritizing bit planes. In the example shown in FIG. 22, the original image 90 (32 × 32 pixels) is converted into four tiles of 16 × 16 pixels, tile 90 _t0 , tile 90 _t1 , and tile 90.
_In the case of division by _t2 and tile 90 _t3 , the size of the precinct and the code block at the decomposition level 1 are 8 × 8 pixels and 4 × 4 pixels, respectively. The numbers of the precincts and the code blocks are assigned in the raster order. In this example, the precincts are assigned numbers 0 to 3, and the code blocks are assigned numbers 0 to 3.
For the pixel expansion outside the tile boundary, a wavelet transform is performed by a reversible (5, 3) integer transform filter using a mirroring method, and a wavelet coefficient value of decomposition level 1 is obtained.

Further, a conceptual diagram showing an example of a typical layer configuration is also shown for the tile 90 _t0 (tile 0) / precinct 90 _p3 (precinct 3) / code block 90 _b3 (code block 3). The converted code block 90 _w3 is the code block 90
_Wavelet conversion of _{b3 is performed} by a reversible (5,3) integer conversion filter to obtain a wavelet coefficient value of decomposition level 1. The converted code block 90 _w3 includes subbands (1LL, 1HL, 1LH,
1HH), and each subband is assigned a wavelet coefficient value.

The layer structure is easy to understand when the wavelet coefficient values are viewed from the horizontal direction (bit plane direction). One layer is composed of an arbitrary number of bit planes. In this example, layers 0, 1, 2, and 3 consist of 1, 3, 1, and 3 bit planes, respectively.
Then, a layer including a bit plane closer to the LSB is subject to quantization first, and conversely, a layer closer to the MSB remains without being quantized to the end. Here, the method of discarding from the layer close to the LSB is called truncation, and it is possible to finely control the quantization rate.

The entropy encoding unit 53 shown in FIG. 18 encodes each component tile by probability estimation from the context and the target bit. In this way, the encoding process is performed on all the components of the original image in tile units.

Finally, the tag processing unit 54 combines all the encoded data from the entropy coder unit into one code stream and performs a process of adding a tag to it. FIG. 23 is a diagram simply showing an example of the structure of a code stream. A tag called a header (a main header 100 and a tile part header 101, respectively) is placed at the beginning of the code stream and the beginning of a partial tile constituting each tile. Information is added, followed by encoded data (bitstream 102) for each tile. Then, the tag is placed again at the end 103 of the code stream.

On the other hand, at the time of decoding, contrary to the time of encoding,
Image data is generated from the code stream of each tile of each component. This will be briefly described with reference to FIG. In this case, the tag processing unit 54 interprets the tag information added to the externally input code stream, decomposes the code stream into code streams for each tile of each component, and Is subjected to a decoding process. The positions of the bits to be decoded are determined in the order based on the tag information in the code stream, and the inverse quantization unit 52
Then, a context is generated from the arrangement of the peripheral bits (already decoded) at the target bit position. The entropy decoding unit 53 generates a target bit by decoding from the context and the code stream by probability estimation, and writes it to the position of the target bit.

Since the data decoded in this manner is spatially divided for each frequency band, the two-dimensional wavelet inverse transform unit 51 performs an inverse two-dimensional wavelet transform on the data to obtain each component of the image data. Are restored. The restored data is converted into the original color system data by the color space inverse converter 50.

In the case of the conventional JPEG compression / expansion format, the tile described in JPEG2000 above may be read as a square block having eight pixels on one side, which is subjected to two-dimensional discrete cosine transform.

Although the description has been given of a general still image, the technique can be extended to a moving image. In other words, each frame of a moving image is composed of one still image, and these still images are displayed at a frame speed optimal for an application to be a moving image. The video data is obtained by continuously encoding the original still image or decoding the compressed still image data, and thus is basically the same as the operation of compressing and expanding the still image. This is a function called motion compression / expansion processing of a still image. This method has a function not available in the MPEG format video file widely used for moving images at present, that is, it has an advantage that a high-quality still image can be edited on a frame basis. It is starting to attract attention in the business field. Eventually,
It has the potential to spread to the general consumer.

The specifications required for the motion still image compression / decompression algorithm differ greatly from the general still image compression / decompression algorithm in the processing speed.
This is because a frame rate that greatly affects the quality of a moving image is determined. Due to the necessity of processing in real time, at the present stage, the implementation method is limited to a method highly dependent on hardware such as an ASIC or a DSP. Eventually, software will be able to perform sufficiently high-speed processing, but by then, it is necessary to wait for further progress in process device technology in the semiconductor field, parallelizing compiler technology in the software field, etc. There seems to be.

[0029]

However, according to the above-mentioned prior art, there is a problem that when the compression / expansion processing is performed under a condition of a high compression ratio, the boundaries between tiles become conspicuous. When the original image to be subjected to the compression / expansion processing has a large spatial area, or when each color component has a deep gradation level, the data amount of the image becomes very large. The idea of tiling was born in order to simultaneously solve the growing demand for high-definition still images in the market and the technical problem of increasing the amount of image data.

If an original image having a very large amount of data is processed without any contrivance, a very large memory area is required to hold a working area for processing image data and a processing result. Thus, the processing time required for compression or decompression is very long. In order to avoid such a situation, it is common practice to divide the original image into rectangular sections, or so-called tiles (called blocks in the case of JPEG), and to perform compression and decompression processing for each area. The idea of dividing the space by these tiles has enabled the increase in memory and process time to a realistic level.

However, the tile division of the original image causes a new problem, that is, the manifestation of the tile boundary described above. This phenomenon occurs when the compressed image data generated by compressing (encoding) the original image in an irreversible (lossy) manner under a condition of a high compression ratio is decoded (decoded) into the original image. In particular, when displaying a high-resolution still image with a large area or a moving image that frequently uses a high compression ratio, even if the image quality in the tile is kept good, the appearance of the tile boundary becomes subjective. Has a great effect on This may cause a serious problem in the future, which is one of the advantages of JPEG2000, namely, that the image quality is less deteriorated even at a high compression ratio.

FIG. 24 is a view showing an example of an image after losslessly compressing and expanding the original image by a factor of 75.
FIG. 3 is a diagram showing an example of an original image and an error image after expansion. Portions indicated by arrows 110a and 111a in FIGS. 24 and 25 are boundaries between tiles adjacent to each other.
A noticeable discontinuous line is observed in this part.

When an image is compressed and decompressed at a high compression rate,
It is considered that the reason why the tile boundaries become clearly visible is in the two-dimensional wavelet transform processing. That is, when the low-pass filter / high-pass filter in the horizontal direction and the low-pass filter / high-pass filter in the vertical direction each perform the filter operation, the operation target region protrudes outside the tile where no image data exists. This ratio increases as the composition level increases.

In the JPEG2000 format, an irreversible (9,
7) A floating point conversion filter and a reversible (5,3) integer conversion filter are recommended, respectively. Here, using the latter (5, 3) reversible wavelet filter as an example, the detailed operation of the wavelet transform will be specifically described, and the reason why a tile boundary appears will be described.

FIG. 26 is a diagram showing an example of pixel expansion using the mirroring method. As shown in FIG. 26, a case is considered where characters “RICOH” are arranged in a certain column of a tile 112 of interest. Each character corresponds to one pixel value, the first character “R” is the k-th character, and the last character “H” is m
It is assumed that the pixel is the nth pixel. When performing a wavelet transform on this tile 112, the pixel before the k-th pixel and m
Since several pixels after the first pixel are required, as shown in FIG.
Need to extend the pixel outside of. Pixel 113 indicates an expanded pixel.

In the (5, 3) reversible wavelet filter, the value of the odd-numbered pixel and the value of the wavelet coefficient of the even-numbered pixel are calculated by the following equations (1) and (2), respectively. Here, C (2i + 1) and C (2i) are wavelet coefficient values, while P (2i + 1) and P (2i) represent pixel values. C (2i + 1) = P (2i + 1)-| _ (P (2i) + P (2i + 2)) / 2_ |, for k-1 <= 2k + 1 <m + 1 Equation (1) C (2i) = P (2i ) + | _ (C (2I−1) + C (2I + 1) +2) / 4_ |, k <= 2I <m + 1 (2)

FIG. 27 shows pixel values and wavelet coefficient values at decomposition level 1 when lossless (5,3) wavelet transform is performed on a square tile of 16 pixels × 16 pixels. FIG. The numbers arranged outside the tiles shown in FIG. 27A are pixel values expanded by the mirroring method. Then, FIG. 28 is based on the wavelet coefficient derived in FIG.
It is a figure showing an example of a pixel value of a square tile of 16 pixels x 16 pixels obtained by performing inverse transformation.

A tile having a pixel value shown in FIG. 27A is provided with a vertical high-pass filter shown in FIG.
A vertical low-pass filter shown in FIG. Next, by applying a horizontal low-pass filter and a horizontal high-pass filter to the result of applying the vertical low-pass filter of FIG. 27C, the LL component of the wavelet coefficient at the decomposition level 1 (see FIG. 27D )) And the HL component (FIG. 27E) are obtained. On the other hand, by applying a horizontal low-pass filter and a horizontal high-pass filter to the result of applying the vertical high-pass filter of FIG. 27B, the LH component of the wavelet coefficient at the decomposition level 1 (FIG. 27F) ) And the HH component (FIG. 27 (G)) are obtained.

FIG. 28 shows 16 pixels × 1 obtained by performing inverse transform from the wavelet coefficients derived in FIG.
It is a figure showing an example of a pixel value of a 6 pixel square tile.
FIG. 28A shows interleaved arrangement of coefficient values of each subband of decomposition level 1 obtained by the forward wavelet transform described with reference to FIG.

FIG. 28B shows a horizontal inverse transform filter for odd-numbered pixels for these coefficient values.
FIG. 28 (C) shows the result of applying a horizontal inverse transform filter to an even pixel, and FIG. 28 (C) shows the result of applying a vertical inverse transform filter to an even pixel subsequently to an odd pixel. .

FIG. 29 compares the pixel values of the original image shown in FIG. 27 (A) with the pixel values obtained by lossless conversion / inversion by the pixel expansion by the mirroring method shown in FIG. 28 (C). FIG. 9 is a diagram showing an example of the result obtained. here,
The error is represented by a difference between individual pixels. It can be seen that, for all tiles, the pixel values after the compression / expansion processing completely match the pixel values of the original image.

FIG. 30 shows the pixel values and the wavelet coefficient values at the decomposition level 1 when the (5,3) lossless wavelet transform is executed in a lossy manner on a square tile of 16 pixels × 16 pixels. FIG. The numbers arranged outside the tiles shown in FIG. 30A are pixel values expanded by the mirroring method. However, the coefficient values show the results of performing quantization and inverse quantization for easy comparison with FIG. 30D shows the LL component, FIG. 30E shows the HL component, and FIG.
The H component is shown, and the HH component is shown in FIG. In this example, the quantization step size is 4 (L
L component) / 32 (HL & LH component) / 64 (HH component)
It is. The quantized wavelet coefficient value is a value obtained by dividing the coefficient of each subband by the quantization step size and adding the original sign to the floor function.

FIG. 31 shows an example of pixel values of a square tile of 16 pixels × 16 pixels obtained by performing an inverse transform from the quantized and dequantized wavelet coefficients derived in FIG. FIG. The forward and backward wavelet transforms are the same as in the lossless case, and a detailed description will be omitted.

FIG. 32 compares the pixel values of the original image shown in FIG. 30 (A) with the pixel values obtained by lossy conversion / inversion by the pixel expansion by the mirroring method shown in FIG. 31 (C). FIG. 9 is a diagram showing an example of the result obtained. According to this example, an error occurs unlike the case of lossless (quantization step size = 1). In particular, a large error is seen in the pixel value near the tile boundary. This is why the tile boundaries are visually noticeable at low bit rates.

Conventionally, in order to solve such a problem, it has been proposed to use image data of adjacent tiles, that is, to overlap boundaries between adjacent tiles. (In the JPEG2000 baseline, adjacent tile boundaries are not allowed to overlap.) Also, a so-called post, which makes the resulting tile boundaries inconspicuous with a completely different image processing algorithm, is used. By filtering, etc.
Was compatible.

The present invention has been made in view of the above points, and provides a new solution different from the above-described tile overlap, post-filter processing, and the like. That is, under high compression ratio conditions, image data obtained by lossy compression of an original image, image compression devices capable of significantly reducing the appearance of tile boundaries that occur when decoding the original image, It is an object of the present invention to provide a decompression device, an image compression method, an image decompression method, a program that can be executed by a computer, and a recording medium storing the program.

[0047]

In order to achieve the above object, according to the first aspect of the present invention, in order to keep the quantization rate near the boundary between adjacent tiles lower than in other areas, the quantization rate is set to be relatively small. The bit plane in the area near the tile boundary that keeps the quantization low is shifted to the upper layer side with respect to the bit plane in the area far from the tile boundary that keeps the quantization ratio relatively high, so that the layers are different. I do. Further, in the invention of claims 2, 3, 4, and 5, means for controlling a quantization rate of a pixel near a tile boundary different from the above,
That is, it is designated by the selective area processing (ROI). According to the sixth and seventh aspects of the present invention, as still another means, the coefficient value before quantization of an area where the quantization rate is kept low is held.

As described above, the tile boundaries are not noticeable even at a high compression ratio without performing the overlap between adjacent tiles or performing post-filter processing as in the related art.

According to the eighth and ninth aspects of the present invention, when the bit plane is reconstructed into different layers, the correlation between the code block and the precinct tile in the coefficient domain and the spatial domain is specifically defined. .

According to the tenth to seventeenth aspects of the present invention, the code block and the precinct bit plane which are in contact with the tile boundary are shifted to a higher layer, and the reconstruction level, the sub-band, and the shift are used when reconstructing. Are specifically described.

The eighteenth aspect of the present invention is such that even a single layer can be reconstructed.

The invention of claims 19, 20, and 24 describes expansion of a compressed image compressed by the image compression apparatus according to any one of claims 1 to 18.

According to a twenty-first aspect of the present invention, in order to keep the quantization rate near the boundary between adjacent tiles lower than in other areas, a bit plane in the tile border neighborhood area where the quantization rate is relatively low is defined as: For bit planes in areas away from tile boundaries that keep the quantization rate relatively high,
The step of shifting to a higher layer side to make the layer different is included. Also, the invention of claim 22 includes a step of designating by selective area processing which is a means different from the layer reconstruction. Further, the invention according to claim 23 includes a step of retaining a coefficient value before quantization in a region where the quantization rate is suppressed to be low.

According to a twenty-fifth aspect of the present invention, there is provided an image compression apparatus according to any one of the first to eighteenth aspects, or as each means of the image compression apparatus, or an image decompression apparatus according to the nineteenth or twentieth aspect. Alternatively, as each means of the image decompression device, a program for causing a computer to function and a recording medium on which the program is recorded are defined.

The invention of each claim has the following configuration. According to a first aspect of the present invention, there is provided a tile designating means for designating a rectangular tile of an arbitrary size with respect to an image, an image dividing means for dividing an image using the designated rectangular tile, and a rectangular form obtained by dividing the image. A bit plane decomposing unit that decomposes a tile into bit planes for each tile, a bit plane ranking unit that ranks the decomposed bit planes in accordance with a coding order, and a layer is configured by the ranked bit planes. Layer configuration means, and has a bit plane in a region near a tile boundary that keeps the quantization rate relatively low, and a bit plane in a region distant from the tile boundary that keeps the quantization rate kept relatively high, Shift the layers so that they are included in the upper layer of the layer consisting of the ranked bit planes, and make the layers different It is obtained by the features.

According to a second aspect of the present invention, there is provided a tile designating means for designating a square tile of an arbitrary size with respect to an image, an image dividing means for dividing an image using the designated square tile, Bit plane decomposing means for decomposing the divided square tiles into bit planes for each tile, bit plane prioritizing means for prioritizing the decomposed bit planes according to the encoding order, and And an area for suppressing the quantization rate is specified by selective area processing.

According to a third aspect of the present invention, in the second aspect of the present invention, the target of the selective area processing is a pixel, a code block, or a precinct near a tile boundary. .

According to a fourth aspect of the present invention, in the second aspect of the present invention, the selective area processing is specified for an area near a tile boundary in an image.

According to a fifth aspect of the present invention, in the second aspect of the present invention, the selective area processing is specified by an ROI.

According to a sixth aspect of the present invention, there is provided a tile designating means for designating a rectangular tile of an arbitrary size with respect to an image, an image dividing means for dividing an image using the designated rectangular tile, Bit plane decomposing means for decomposing the divided square tiles into bit planes for each tile, bit plane prioritizing means for prioritizing the decomposed bit planes according to the encoding order, and And a layer configuration unit configured to hold a coefficient value before quantization in an area where the quantization rate is suppressed to be low.

According to a seventh aspect of the present invention, in the sixth aspect of the present invention, the place where the coefficient value before quantization of the region where the quantization rate is suppressed is held in the tile part header. It was done.

According to an eighth aspect of the present invention, in any one of the first, second, and sixth aspects of the present invention, the re-encoding is performed on a layer different from a bit plane in a region apart from a tile boundary for maintaining the quantization rate relatively high. The minimum unit of the set of configured bit planes is a code block, and at least one of the code blocks is not in contact with the tile outer edge.

According to a ninth aspect of the present invention, in any one of the first, second and sixth aspects of the present invention, a layer different from a bit plane in a region distant from a tile boundary for maintaining the quantization rate relatively high is reproduced. The maximum unit of the set of bit planes is a precinct, and at least one of the precincts is not in contact with the outer edge of the tile.

According to a tenth aspect of the present invention, in any one of the first, second, and sixth aspects of the present invention, a layer different from a bit plane in a region apart from a tile boundary for maintaining the quantization rate relatively high is reproduced. The constituted bit plane is characterized in that the decomposition level is 1 or more.

An eleventh aspect of the present invention is characterized in that, in the tenth aspect of the present invention, the processing time and image quality are different depending on the decomposition level.

In a twelfth aspect of the present invention, in any one of the first, second, and sixth aspects of the present invention, a layer different from a bit plane in a region distant from a tile boundary for maintaining the quantization rate relatively high is reproduced. The configured bit plane is characterized in that the number of subbands is one or more.

According to a thirteenth aspect, in the twelfth aspect, the processing time and the image quality are different depending on the number of subbands.

According to a fourteenth aspect of the present invention, in the first or second aspect, the bit plane reconstructed in a different layer from a bit plane in a region apart from a tile boundary for maintaining the quantization rate relatively high is , The wavelet coefficient value of the bit plane is larger than a predetermined value.

According to a fifteenth aspect, in the fourteenth aspect, as the difference between the wavelet coefficient value of the bit plane constituting the code block and a predetermined value is larger, the bit plane is assigned to a higher layer. The feature is that the layers are reconstructed so as to be absorbed.

A sixteenth invention is characterized in that, in the invention of the fourteenth invention, when the wavelet coefficient value is obtained, a mirroring method is used for extended pixel interpolation outside the tile.

According to a seventeenth aspect of the present invention, in any one of the first, second, and sixth aspects of the present invention, when a predetermined number or more of the code blocks constituting the precinct are in contact with the tile boundary, all of the code blocks included in the precinct The bit planes are shifted so as to be included in the upper layer, and the layers are reconfigured.

According to an eighteenth aspect of the present invention, in any one of the first, second, and sixth aspects, the number of layers before reconstruction is:
It is characterized by one or more.

The nineteenth aspect of the present invention relates to the first to eighteenth aspects.
An image decompression device for decompressing a compressed image compressed by the image compression device according to any one of the above, wherein the compressed image is added to information of a square tile specified by the tile specifying means, which is included in the compressed image. It is characterized in that it is expanded on the basis of the above.

According to a twentieth aspect of the present invention, in the nineteenth aspect, an entropy decoder, an inverse quantizer, and a two-dimensional discrete wavelet inverse transformer are provided.

The invention according to claim 21 is a tile specifying step of specifying a square tile of an arbitrary size for an image,
An image dividing step of dividing an image using the designated square tile; a bit plane disassembling step of disassembling the square tile obtained by dividing the image into bit planes for each tile; Bit plane ranking step to rank according to, and having a layer configuration step of configuring a layer by the ranked bit plane, the bit plane of the tile boundary near area to keep the quantization rate relatively low, A bit plane in a region apart from a tile boundary that maintains a relatively high quantization rate is shifted so as to be included in an upper layer side of the layer composed of the ranked bit planes, and the layer is shifted. It is characterized by being different.

The invention according to claim 22 is a tile designating step of designating a rectangular tile with an arbitrary size for an image,
An image dividing step of dividing an image using the designated square tile; a bit plane disassembling step of disassembling the square tile obtained by dividing the image into bit planes for each tile; And a layer configuration step of configuring a layer based on the ranked bit planes, wherein an area for suppressing the quantization rate is specified by the selective area processing. It is a characteristic.

According to a twenty-third aspect of the present invention, a tile designating step of designating a rectangular tile of an arbitrary size for an image,
An image dividing step of dividing an image using the designated square tile; a bit plane disassembling step of disassembling the square tile obtained by dividing the image into bit planes for each tile; And a layer configuration step of configuring a layer based on the ranked bit planes, and retains a coefficient value before quantization in a region where the quantization rate is to be reduced. It is characterized by keeping.

The invention according to claim 24 is the invention according to claims 21 to 2
3. An image decompression method for decompressing a compressed image compressed by the image compression method according to any one of 3., wherein the compressed image includes information on a square tile specified by the tile specifying means and included in the compressed image. It is characterized in that it is expanded on the basis of.

The twenty-fifth aspect of the present invention provides the first to eighteenth aspects.
A program for causing a computer to function as the image compression device according to any one of the claims or each unit of the image compression device, or as the image expansion device according to the claim 19 or 20 or each unit of the image expansion device. It is.

The invention according to claim 26 is the invention according to claims 1 to 18
A program for causing a computer to function as the image compression device according to any one of the claims or each unit of the image compression device, or as the image expansion device according to the claim 19 or 20 or each unit of the image expansion device. Is a computer-readable recording medium having recorded thereon.

[0081]

(First Embodiment) FIG. 1 is a block diagram showing a configuration example of an image compression / decompression apparatus according to an embodiment of the present invention. The inverse transform unit 11 is a first component. The first component 11 includes an interpolation method selecting unit 11a, a wavelet transform / inverse transforming unit 11b, a quantization rate selecting unit 11c, a quantization / inverse quantizing unit 11d, Similarly, the second component 12 includes an interpolation method selecting unit 12a, a wavelet transform / inverse transforming unit 12b, a quantization rate selecting unit 12c, a quantization / inverse quantizing unit 12d, and an entropy encoding / decoding unit 11e. The third component 13 includes an encoding / decoding unit 12e, an interpolation method selecting unit 13a, a wavelet transform / inverse transforming unit 13b, a quantization rate selecting unit 13c, a quantization / inverse quantizing unit 13d, It consists Ntoropi coding and decoding unit 13e, 14 is a code stream processing unit.

Although the image compression / expansion apparatus shown in FIG. 1 shows one conversion / inversion unit in each processing block so as to cope with image compression and expansion, the conversion unit and the inverse conversion unit are used. It may take a form divided into parts.

The image compression / decompression device described above is intended for still images. Processing is for each component 1
It proceeds in parallel for each of 1, 12, and 13. The color space here is RGB or YUV. Hereinafter, the case of compression will be described using the first component 11 as an example. The interpolation method selection unit 11a selects an interpolation method for calculating a pixel value or a wavelet coefficient value extended outside the tile boundary. The interpolation method includes a mirroring method, a point-symmetric interpolation method, a method using actual neighboring tile pixel values, and the like. Here, the mirroring method is selected. Subsequently, the wavelet transform / inverse transform unit 11b performs a wavelet transform on the original image. Then, the quantization rate selection unit 11c creates a new layer for the bit plane of the code block in contact with the tile boundary among the wavelet coefficient values obtained by the mirroring method. Specifically, more young layer numbers, for example, layers 0, 1, 2,..., Are reassigned to the bit planes so that even if truncation is performed, data is not discarded.

FIGS. 2 and 3 show an original consisting of four tiles.
What happens when the wavelet transform is applied to the image
FIG. In FIG. 2, the original image 20 has four tags.
Divided into 20_t0Are tiles 0, 20_t1Is
Tile 1, 20_t2Is tile 2, 20_t3The tile
3, for example, tile 20_t3Is 1HL,
1 LH, 1 HH, each showing the first precinct
Think 20_p1The precinct 20_p1Is
Code block 20 indicating the ninth code block_b9
including. Also, in FIG. 3, the original image 21 has four tags.
Divided into 21_t0Are tiles 0, 21_t1Is
Tile 1, 21_t2Is tile 2, 21 _t3The tile
3, for example, tile 21_t3Is 2HL,
2LH, 2HH, 8th precinct
Think 21_p8The precinct 21_p8Is
Code block 21 indicating the ninth code block_b9
including. In this example, tile 0 and the upper left corner of the original image
Match, but there can be an offset.

FIG. 2 shows the composition level 1 and FIG. 3 shows the composition level 2.
The relationship between the precinct and the code block is shown. In this example, the number of tiles = 4, the number of precincts =
9, the number of code blocks = 12. The size of the precinct and code block can be specified for each decomposition. Here, for ease of explanation, the same number of precincts and the same number of code blocks are used for both decomposition level 1 and decomposition level 2. In FIG. 2 and FIG. 3, the colored blocks are the code blocks of the HL / LH / HH subbands that are in contact with the tile boundaries, and are the portions to be subjected to layer reconstruction described later.

FIG. 4 is a diagram showing the relationship between the precinct of each subband and the code block in the case of the decomposition level 1 for the tiles 20 _t0 to 20 _t3 shown in FIG. FIG. 4A shows the tile 20.
_t0 , FIG. 4 (B) shows tile 20 _t1 , and FIG. 4 (C) shows
The tile 20 _t2 and FIG. 4D show the case of the tile 20 _t3 , respectively. The vertical axis indicates the precinct number, and the horizontal axis indicates the code block number. In the figure, the colored portions are the code blocks that are in contact with the tile boundaries.
For example, in the tile 20 _t0 shown in FIG. 4A, the precinct number 7 indicates that the second, third, sixth, seventh, tenth and eleventh code blocks are in contact with the tile boundary, and the precinct number is Since the number 4 is at a position not in contact with the outer edge of the tile, it can be seen that the code block included in any tile does not touch the tile boundary.

FIG. 5 is a diagram showing an example of a state before and after layer reconstruction to which the present invention is applied. FIG. 5B shows a layer before reconstruction, and FIG. 5A shows a layer after reconstruction. Before the reconstruction shown in FIG. 5B, 2HL · 2
The same layer is formed by LH / 2HH / 1HL and 1LH / 1HH. In this example, similarly to the example shown in FIG. 4A, the decomposition level 1, the tile 20 _t0 (tile 0), and the precinct number 7 are taken up.

The vertical axis shown in FIG. 5 is a 12-bit wavelet coefficient, MSB at the top and LSB at the bottom.
Is assigned. The horizontal axis is the code block number, from the right 1HH (code block numbers 8, 9, 1).
0, 11), 1LH (code block number 4,
5, 6, 7), 1HL (comprising code block numbers 0, 1, 2, 3), hereinafter 2HH, 2LH, 2H
They are arranged in the order of L subbands. A layer is composed of bit planes of each code block, and layers 0, 1,..., 9 are formed in order from the top. The higher the layer and the deeper the composition level, the more important data to be left (data not to be discarded). Here, as shown in FIG. 5 (B), the bit plane layers are configured in a stepwise manner. For example, when the layer 9 is discarded by truncation as a layer including a bit plane close to the LSB, 1
Four bits are discarded in the HH code block and one bit is discarded in the 2HL code block. In FIG. 5, a white line drawn between the layer 9 and the layer 8 indicates a boundary with the layer 9 discarded at the time of truncation for explanation.

Since the code blocks of code block numbers 2, 3; 6, 7, 10 and 11 shown in gray in FIG. 5B are in contact with the tile boundaries, after the layer is reconstructed, the code blocks of FIG. ) Are included in a higher layer, for example, layer 0, with respect to the bit planes of other code blocks 0, 1: 4, 5; Has been shifted. This means that when the wavelet coefficient value is discarded by truncation, the quantization rate can be suppressed to one fourth as compared with the portion not in contact with the boundary. This is, for example, shown in FIG.
In the case of the 1HH code blocks 10 and 11 shown in FIG. 5, when the layer 9 is discarded, it is 4 bits, that is, the quantization rate is 16. However, the 1HH code block after reconfiguration shown in FIG. In layers 10 and 11, when layer 9 is discarded, 2 bits, that is, a quantization rate of 4
Which means that the quantization rate becomes 1/4.

The layer for shifting the bit plane in the area near the tile boundary is not limited to layer 0, but may be a higher layer, for example, layer 1 or layer 2
May be included. In any case, shifting to the upper layer prevents data from being discarded during truncation. After the quantization rate in the vicinity of the tile boundary is determined in this way, the wavelet coefficients are converted into the quantization units 11d and 12 shown in FIG.
d, 13d, and are further encoded by entropy encoding units 11e, 12e, 13e. Finally,
The code stream processing unit 14 shown in FIG. 1 generates a code stream.

In the above example, a plurality of layers are configured. However, the processing can be performed by regarding the plurality of layers as one layer. For example, considering layers 0 to 8 shown in FIG. 5B as one layer,
The layer after reconstruction shown in FIG. 5A shifts the bit plane in contact with the tile boundary so as to be included in the bit plane upper side with respect to the bit plane distant from the tile boundary, thereby lowering the bit plane. The layer is reconstructed in such a manner that the bit planes (the bit planes of layers 7 and 8 in this example) are added to the original layer.

In the above embodiment, the layer is reconstructed only in the case of the decomposition level 1, but the case of the decomposition level 2 can be similarly considered. FIG. 6 is a diagram illustrating an example of layer reconstruction in the case of the decomposition level 2. As in the example shown in FIG. 5, the tile 20 _t0 and the precinct number 7 are shown. Here, the code blocks in the second, third, sixth, seventh, tenth, and eleventh code blocks adjacent to the tile boundary are operated so as to be included in the next higher layer. That is, the quantization rate can be reduced to half of that of the portion not in contact with the boundary.

FIG. 7 is a diagram showing another embodiment of the layer reconstruction. In this embodiment, the layer is reconstructed only when the wavelet coefficient value near the tile boundary obtained by the interpolation method is larger than a predetermined value.
This is based on the empirical fact that when a large distortion appears at the tile boundary, the wavelet coefficient value near the boundary increases. 1HL subband code block 2
Although the code block 3 and the code block 3 are in contact with the tile boundary, the layer reconstruction is not performed in the 1HL sub-band because the wavelet coefficient value is smaller than a predetermined value. On the other hand, for both the 1LH and 1HH subbands, the layer is reconstructed because the wavelet coefficient values are larger than the values specified in advance.

Further, when the wavelet coefficient value near the tile boundary obtained by the interpolation method is larger than a predetermined value, it is considered that the larger the difference, the larger the tile boundary distortion corresponding to the difference. Therefore, by determining the shift amount of the bit plane while corresponding to the magnitude of the difference, it is possible to more easily control the quantization rate.

FIG. 8 is a diagram showing an example in which the present invention is applied to a still image. In FIG. 8, reference numeral 22 denotes a first component,
Reference numeral 23 denotes a second component, and reference numeral 24 denotes a third component. For example, the first component 22 has a tile boundary 22a, a portion 22b in which the quantization rate is kept low, and a portion 22c in which the quantization rate is not controlled. And the state where the quantization rate in the vicinity of the tile boundary is suppressed. FIG. 9 is a diagram showing an example in which the present invention is applied to a moving image. In each frame (25, 26, 27), quantization rate control near a tile boundary is performed. 8 and 9
In FIG. 7, the shaded portion is a region where the quantization rate is suppressed to a low level, and the portion where there is no shadow despite the tile boundary is the wavelet coefficient value near the tile boundary as in the embodiment shown in FIG. Is less than a pre-specified value,
This is because there was no need to perform layer reconstruction. That is, the portion where the tile boundary was originally not noticeable.

In the above embodiment, the layer is reconstructed in units of code blocks. In this case, it is possible to set an optimal quantization rate that minimizes tile boundary distortion. On the other hand, when the original image size is large, the processing time may increase. When high-speed processing is prioritized as in moving image display, layer reconstruction may be performed in units of precincts. This may not be possible to select the optimal quantization rate, but it can be processed in real time There is an advantage.

FIG. 10 is a diagram showing a result of reconstructing all the code blocks included in the precinct so as to be included in the upper layer. According to FIG. 10, when a predetermined number or more, for example, half or more code blocks in the precinct are in contact with the tile boundary, all the code blocks included in the precinct are reconfigured to be included in the upper layer. The results are shown. According to this, the entire sub-band of 1HL, 1LH, 1HH is uniformly absorbed in the upper layer.

By the way, lowering the quantization rate near the tile boundary in order to reduce the tile boundary distortion results in lowering the image compression rate. Therefore, in order to alleviate the disadvantage, a method of devising a geometric relationship between a precinct or a code block and a tile is considered. That is, if a precinct or a code block that does not touch the tile boundary is created, the quantization rate there may be higher than that of the tile boundary, and this contributes to the suppression of the increase in the compression rate of the entire image.

FIGS. 11 and 12 show specific examples of the code block and the precinct, respectively. FIG.
1 is a diagram showing a specific example for the code block, a tile 28 _t is divided into precincts 28 _p, the precinct 28 _p, the code block 28 _b is divided into the fourth code block 28 therein _{Since b4} does not touch the outer edge of the tile, the quantization rate can be higher than that of the tile boundary. FIG. 12 is a diagram illustrating a specific example of the precinct, in which the tile 29 _t is the precinct 29.
is divided into _p, the precinct 29 _p is 4 th precinct 29 _p4 therein, since no contact with the tile edge portion can be increased as compared with quantization index to the tile boundaries.

Further, the compression effect differs depending on the decomposition level of the bit plane reconstructed into different layers and the number of subbands. For example, if the decomposition level is 1, the processing time can be shortened, but the image quality is somewhat coarse. When the decomposition level is set to 2 or more, the image quality is improved but the processing time is increased. The same can be said for the number of subbands to be divided. These settings are appropriately set according to the use environment, the purpose of use, and the like.

(Second Embodiment) The first embodiment described above is an example in which layers are reconstructed.
Selective region processing (ROI: Region of Int)
est), there is a method of controlling the quantization rate near the tile boundary to improve the tile boundary distortion. JPEG2
With an extended specification of 000 (although not fixed at the time of this application), it is expected that the quantization rate can be reduced uniformly for an arbitrary shape. This means that the bit plane for the ROI is set to 1 separately from the wavelet coefficient value.
A method for easily performing the quantization rate control by preparing bits and setting bits of this plane at pixel positions that are to be treated as an ROI, and filling this bit plane with 0 at other pixel positions It is.

FIG. 13 is a diagram showing an embodiment of the selective area processing by the ROI.
0 _t0 , 30 _t1 , 30 _t2 , 30 _t3 , and a cross-shaped ROI having a predetermined width along each tile boundary
31 is specified. In FIG. 13A, the ROI 31 is specified with a width of 8 pixels. The wavelet coefficient value generated from the pixel values existing in this cross-shaped area is
Based on the value specified in advance, the quantization rate is reduced. As a result, the distortion at the tile boundary is improved. Here, the detailed description thereof is omitted. Further, FIG. 13B shows that the RO is added to the tile boundary region in the image.
An example in which I31 is set is shown. In this example, ROI 31 is used for the tile boundary area inside the image belonging to one frame.
Is set.

(Third Embodiment) As a more thorough method, a wavelet coefficient value near a tile boundary before quantization is held separately from a coefficient value after quantization. Conceivable. In JPEG2000,
To insert a comment at a predetermined position in the code, COM
Markers are defined. Any number of COM markers can be inserted into the main header or tile part header, and arbitrary data can be inserted up to 65535 bytes following the marker. Therefore,
In the present embodiment, a COM marker is inserted into the tile part header, and after this marker, 1H that is in contact with the tile boundary is inserted.
Wavelet coefficient values of the L, 1LH, and 1HH pixels before quantization are inserted in a predetermined format.

FIG. 14 shows 1HL, 1 in contact with the tile boundary.
LH, a diagram showing an example of a format defined in advance wavelet coefficients before quantization of pixels 1HH, the original image 32 is divided into four tiles, the tiles _{32 t0,} the tiles _{32 t1,} the tiles _{32 t 2} consists tile _{32 t3} Prefecture, region 33 colors with a has 1HL in contact with the boundary of the tile _{32 t0,} 1LH, wavelet coefficients prior to quantization of pixels of 1HH. And for simplicity,
As shown in FIG. 14, the coefficients on the four sides of each sub-band of 1HL, 1LH, and 1HH are expressed as: upper side → right side → lower side → left side.
The data is stored clockwise and in the order of 1HL → 1LH → 1HH with a fixed length of 2 bytes per coefficient.

Since the contents of the COM marker, that is, the order (format) of writing as described above are free, they generally have no meaning for a third-party decoder. That is, decoding is usually performed based on the quantized coefficients present in the code. However, if the content of the COM marker can be known, the quantized value present in the code is not used for the wavelet coefficient value at the tile boundary, and the accurate value embedded in the COM marker is used. It can be used to decode.

In JPEG2000, 9 × 7 and 5 × 3
The use of wavelet transform filters is specified.
When these filters are used, in order to obtain one pixel value by the inverse wavelet transform, a maximum of the wavelet coefficient values of the surrounding 81 pixels and 25 pixels are required, respectively. Further, each coefficient value contributes to derivation of many pixel values. Therefore, the accurate wavelet coefficient value of the tile boundary read from the COM marker is valuable data for the inverse transformation when deriving the pixel value existing at the tile boundary. As a result, the quantization error of the pixel value near the tile boundary can be reduced, and the tile boundary can be made inconspicuous.

FIG. 15 is a flowchart for explaining an image compression method according to an embodiment of the present invention. The image compression method according to the present embodiment is a method for the JPEG2000 system. First, an image is converted into a color space (step S1), and then a square tile of an arbitrary size is specified for the image, and the specified image is specified. The image is divided using the rectangular tiles (step S2). After the tile division, the tiles are further divided into code blocks, and DWT conversion is performed on the divided code blocks (step S3). The code block after DWT is decomposed into bit planes, and MSB to LSB according to the coding order of the decomposed bit plane.
Rank up to. The ordered bit planes make up a layer that is ranked from the highest layer to the lowest layer, and the bit plane in the area near the tile boundary that keeps the quantization rate relatively low A bit plane in a region away from the tile boundary maintained at a high level is shifted so as to be included in the upper layer side of the ranked layers, and is reconfigured into a different layer and quantization is performed (step S4). ). Entropy coding is performed from the upper side of the quantized layer (step S5), and tag processing is performed (step S6).

Here, as a function of JPEG2000, an area in which the quantization rate is suppressed to be low can be designated by selective area processing. It is also possible to hold a coefficient value before quantization in a region where the quantization rate is kept low, and refer to the value when decompressing.

FIG. 16 is a flowchart for explaining an image decompression method according to an embodiment of the present invention. First, tag information is interpreted based on tile division information included in the compressed image (step S11), entropy decoding (step S12), inverse quantization (step S13),
The inverse DWT (step S14) is performed to restore the tiles (step S15), and the restored tiles are integrated to perform the inverse color space conversion (step S16) to generate an expanded image.

The embodiments of the present invention have been described centering on the image compression apparatus, the image decompression apparatus, the image compression method, and the image decompression method of the present invention. The present invention can also be implemented as a program for causing it to function as each means or for causing a computer to execute these methods, or as a computer-readable recording medium on which the program is recorded.

An embodiment of a recording medium storing programs and data for realizing the image compression and decompression functions according to the present invention will be described. As a recording medium, specifically,
CD-ROM, magneto-optical disk, DVD-ROM, floppy (registered trademark) disk, flash memory, memory card, memory stick, and various other ROMs and Rs
AM and the like can be assumed, and by causing a computer to execute the functions according to the above-described embodiments of the present invention on these recording media and recording and distributing a program for realizing the image compression and decompression functions, Facilitates the realization of Then, the program is read by the information processing apparatus by attaching the recording medium as described above to an information processing apparatus such as a computer, or the program is stored in a storage medium provided in the information processing apparatus, and if necessary, By reading, the functions of image compression and expansion according to the present invention can be executed.

[0112]

According to the first or twenty-first aspect of the present invention, the bit plane in the area near the tile boundary where the quantization rate is kept relatively low is set in the area far from the tile boundary where the quantization rate is kept relatively high. By shifting the bit plane so that it is included in the upper layer side and making the layers different, even if the image is compressed at a high compression ratio, the tile boundaries can be recognized in the expanded image as before. No more.

Any one of claims 2, 3, 4, 5, and 22
According to the invention, the region in which the quantization rate is suppressed to be low is specified by the selective region processing, so that the deterioration of the image quality due to the appearance of the tile boundary can be improved more easily.

The invention according to any one of claims 6, 7 and 23 provides still another means, and stores the coefficient value before quantization in the region where the quantization rate is kept low. Even if the image is compressed at a high compression rate, the tile boundary is not visually recognized in the expanded image as in the related art.

According to any one of the first, second, and sixth aspects of the present invention, tile boundary distortion is suppressed even at a high compression ratio without performing overlap between adjacent tiles or performing post-filter processing as in the related art. be able to.

According to the eighth or ninth aspect of the present invention, when the bit plane is reconstructed into different layers, the correlation between the code blocks and the precinct tiles in the coefficient domain and the spatial domain is specifically defined. The size of the original image (VGA, HDTV
Etc.), effectively reducing tile boundary distortion even under various constraint conditions such as processing device performance (memory capacity, processing speed, etc.), and specifications (processing speed, image quality, cost, etc.) required by applications. Can be.

According to any one of the tenth to seventeenth aspects of the present invention, a code block and a precinct bit plane that are in contact with a tile boundary are shifted to a higher layer and recomposed, and
Since the sub-bands, the method of shifting, and the like are specifically described, it is possible to control the quantization rate optimally and finely.

According to the eighteenth aspect of the present invention, the quantization ratio can be easily controlled by enabling reconfiguration even from one layer.

According to any one of the nineteenth, twentieth and twenty-fourth aspects, it is possible to decompress a compressed image that has been compressed by the image compression device according to any one of the first to eighteenth aspects.

According to the invention of claim 25 or 26, the program having the function of the image compression / decompression device according to any one of claims 1 to 20 can be executed, and by executing the program, a high compression ratio can be obtained. In this case, it is possible to suppress a decrease in image quality due to tile division.

[Brief description of the drawings]

FIG. 1 is a block diagram illustrating a configuration example of an image compression / decompression device according to an embodiment of the present invention.

FIG. 2 is a diagram illustrating a state where a wavelet transform is performed on an original image including four tiles.

FIG. 3 is a diagram illustrating a state in which a wavelet transform is performed on an original image including four tiles.

FIG. 4 is a diagram showing a relationship between a precinct and a code block of each subband in the case of a decomposition level 1 for tiles 0 to 3 shown in FIG. 2;

FIG. 5 is a diagram illustrating an example of a state before and after layer reconstruction to which the present invention is applied.

FIG. 6 is a diagram showing an example of layer reconstruction in the case of decomposition level 2.

FIG. 7 is a diagram showing another embodiment of the layer reconstruction.

FIG. 8 is a diagram showing an example in which the present invention is applied to a still image.

FIG. 9 is a diagram showing an example in which the present invention is applied to a moving image.

FIG. 10 is a diagram showing a result of reconstructing all code blocks included in a precinct so as to be included in an upper layer.

FIG. 11 is a diagram showing a specific example of a code block and a precinct.

FIG. 12 is a diagram illustrating a specific example of a code block and a precinct.

FIG. 13 is a diagram showing an embodiment of selective area processing by ROI.

FIG. 14: 1HL, 1LH, 1H in contact with a tile boundary
FIG. 11 is a diagram illustrating an example of a format in which a wavelet coefficient value of an H pixel before quantization is specified in advance.

FIG. 15 is a flowchart illustrating an image compression method according to an embodiment of the present invention.

FIG. 16 is a flowchart illustrating an image decompression method according to an embodiment of the present invention.

FIG. 17 is a block diagram for explaining the basics of the JPEG algorithm.

FIG. 18 is a block diagram for explaining the basics of the JPEG2000 algorithm.

FIG. 19 is a diagram illustrating an example of subbands at each decomposition level when the decomposition level is 3.

FIG. 20 is a diagram illustrating an example of each component of a tiled color image.

FIG. 21 is a diagram illustrating an example of a relationship between a precinct and a code block.

FIG. 22 is a diagram illustrating an example of a procedure for ranking bit planes.

FIG. 23 is a diagram simply showing an example of a structure of a code stream.

FIG. 24 Lossless compression of the original image by a factor of 75
FIG. 4 is a diagram illustrating an example of an image after being expanded.

FIG. 25 is a diagram showing an example of an original image and an error image after expansion.

FIG. 26 is a diagram illustrating an example of pixel extension using a mirroring method.

FIG. 27 shows a pixel value and a decomposition level 1 when a lossless (5,3) wavelet transform is performed on a square tile of 16 pixels × 16 pixels.
It is a figure which shows the wavelet coefficient value at the time of.

28 is a diagram illustrating an example of pixel values of a 16 × 16 pixel square tile obtained by performing an inverse transform from the wavelet coefficients derived in FIG. 27;

29 shows the pixel values of the original image shown in FIG. 27 and FIG.
FIG. 13 is a diagram showing an example of a result obtained by comparing pixel values obtained by performing a lossless conversion / inverse conversion by pixel expansion by the mirroring method shown in FIG.

FIG. 30 shows a pixel value and a decomposition level 1 when a (5, 3) lossless wavelet transform is executed in a lossy manner on a square tile of 16 pixels × 16 pixels.
It is a figure which shows the wavelet coefficient value at the time of.

31 is obtained by performing an inverse transform from the wavelet coefficients after the quantization and the inverse quantization derived in FIG.
It is a figure which shows an example of the pixel value of a square tile of 6 pixels x 16 pixels.

FIG. 32 shows the pixel values of the original image shown in FIG.
FIG. 11 is a diagram showing an example of a result obtained by comparing pixel values obtained by performing conversion and inverse conversion by lossy by pixel expansion by the mirroring method shown in FIG.

[Explanation of symbols]

 10: color space conversion / inverse conversion unit, 11, 22, 70 ... first
Component, 12, 23, 71 ... second component
, 13, 24, 72 ... third component, 11a,
12a, 13a... Interpolation method selection section, 11b, 12b, 13
b: Wavelet transform / inverse transform unit, 11c, 12c,
13c: quantization rate selection unit, 11d, 12d, 13d: quantity
Quantization / inverse quantization unit, 11e, 12e, 13e ... entro
P code / decoding section, 14 code stream processing section,
20, 21, 30, 32, 80, 90: Original image, 20
_t0~ 20_t3, 21_t0~ 21_t3, 28_t, 2
9_t, 30 _t0~ 30_t3, 32_t0~ 32_t3, 70
_t, 71_t, 72_t, 80_t0~ 80_t3, 90_t0~
90_t3, 112 ... tile, 20_p1, 21_p8, 28
_p, 29_p, 80_p4, 80_p6, 90_p0~ 90_p3
... Precinct, 20_b9, 21_b9, 28_b, 80
_b1, 90_b3, 90_w3... code block, 25 ... f
Frames 1, 26 ... Frame 2, 27 ... Frames 3, 28
_b4… Code block that does not touch the outer edge of the tile, 29
_p4... Precinct that does not touch the outer edge of the tile, 31 ... R
OI, 33: tile boundary area, 40, 50: color space conversion
・ Inverse transform unit, 41: discrete cosine transform / inverse transform unit, 4
2, 52 ... quantization / inverse quantization unit, 43, 53 ... entro
P encoding / decoding unit, 51 ... two-dimensional wavelet transformation
Conversion / inverse conversion unit, 54: tag processing unit, 60: decomposition
Tile at the 0th level, 61 ... Decomposition level
1 tile, 62 ... Decomposition level 2 tie
, 63 ... Decomposition level 3 tile, 100
... Main header, 101 ... Tile part header, 102
... bit stream, 103 ... code stream end,
110: expanded image, 111: error image, 113: expanded
Pixel.

 ──────────────────────────────────────────────────の Continued on the front page (72) Inventor Taku Kodama 1-3-6 Nakamagome, Ota-ku, Tokyo F-term in Ricoh Co., Ltd. (reference) 5C059 KK01 LC00 MA00 MA24 MA35 MC11 MC38 ME01 PP01 PP15 PP16 SS15 SS20 SS21 SS28 TA39 TA46 TB00 TC31 UA02 UA05 5C078 AA04 BA53 CA22 DA00 DA01 DA02 DB07 DB19 EA00 EA01 5J064 AA01 BA09 BA16 BB04 BC16 BD03 BD04

Claims (26)

[Claims] 1. A tile designating means for designating a square tile of an arbitrary size with respect to an image, an image dividing means for dividing an image using the designated square tile, and a square tile obtained by dividing the image. Bit plane decomposing means for decomposing into bit planes for each tile, bit plane prioritizing means for prioritizing the decomposed bit planes in accordance with the coding order, and a layer configuration for forming a layer by the bit planes thus ordered Means for ranking the bit planes in the area near the tile boundary that keeps the quantization rate relatively low with respect to the bit planes in the area far from the tile boundary that keeps the quantization rate relatively high. It is characterized by shifting so that it is included in the upper layer side of the layer consisting of Image compression device. 2. A tile designating means for designating a square tile of an arbitrary size with respect to an image, an image dividing means for dividing an image using the designated square tile, and a square tile obtained by dividing the image. Bit plane decomposing means for decomposing into bit planes for each tile, bit plane prioritizing means for prioritizing the decomposed bit planes in accordance with the coding order, and a layer configuration for forming a layer by the bit planes thus ordered Means for specifying an area in which the quantization rate is to be kept low by selective area processing. 3. The image compression apparatus according to claim 2, wherein the target of the selective area processing is a pixel, a code block, or a precinct near a tile boundary. . 4. The image compression apparatus according to claim 2, wherein the selective area processing is specified for an area near a tile boundary in an image. 5. The image compression apparatus according to claim 2, wherein said selective area processing is specified by a ROI. 6. A tile designating means for designating a square tile of an arbitrary size with respect to an image, an image dividing means for dividing an image using the designated square tile, and a square tile obtained by dividing the image. Bit plane decomposing means for decomposing into bit planes for each tile, bit plane prioritizing means for prioritizing the decomposed bit planes in accordance with the encoding order, and a layer configuration for forming a layer by the bit planes thus ordered Means for storing a coefficient value before quantization in a region where the quantization rate is suppressed to be low. 7. The image compression apparatus according to claim 6, wherein the place where the coefficient value before quantization of the region where the quantization rate is suppressed is held in the tile part header. Image compression device. 8. The image compression apparatus according to claim 1, wherein the quantization ratio is reconstructed into a layer different from a bit plane in a region apart from a tile boundary for maintaining a relatively high quantization rate. The minimum unit of a set of bit planes to be processed is a code block, and at least one of the code blocks is not in contact with an outer edge of the tile. 9. The image compression apparatus according to claim 1, wherein the quantization ratio is maintained in a layer different from a bit plane in a region apart from a tile boundary for maintaining a relatively high quantization rate. The maximum unit of a set of bit planes to be processed is a precinct, and at least one of the precincts does not touch the outer edge of the tile. 10. The image compression apparatus according to claim 1, wherein the quantization ratio is maintained in a layer different from a bit plane in a region apart from a tile boundary for maintaining a relatively high quantization rate. An image compression apparatus characterized in that the bit plane to be processed has a decomposition level of 1 or 2 or more. 11. The image compression apparatus according to claim 10, wherein the processing time and the image quality are different according to the decomposition level. 12. The image compression apparatus according to claim 1, wherein the quantization ratio is reconstructed into a layer different from a bit plane in a region apart from a tile boundary for maintaining a relatively high quantization rate. An image compression apparatus, wherein the number of subbands is one or more. 13. The image compression apparatus according to claim 12, wherein the processing time and the image quality are different according to the number of subbands. 14. The image compression apparatus according to claim 1, wherein the bit plane reconstructed into a layer different from a bit plane in a region apart from a tile boundary that maintains the quantization rate relatively high is: An image compression apparatus, wherein a wavelet coefficient value of the bit plane is larger than a value specified in advance. 15. The image compression apparatus according to claim 14, wherein the larger the difference between the wavelet coefficient value of the bit plane forming the code block and the predetermined value, the more the bit plane is absorbed by a higher layer. An image compression apparatus characterized in that a layer is reconstructed in such a manner as to be performed. 16. The image compression apparatus according to claim 14, wherein when the wavelet coefficient value is obtained, a mirroring method is used for extended pixel interpolation outside the tile. 17. The image compression apparatus according to claim 1, wherein when a predetermined number or more of code blocks constituting a precinct contact a tile boundary portion,
An image compression apparatus for reconstructing layers by shifting all bit planes included in the precinct so as to be included in an upper layer. 18. The image compression apparatus according to claim 1, wherein the number of layers before reconstruction is one.
Or, a plurality of image compression devices. 19. An image decompression device for decompressing a compressed image compressed by the image compression device according to claim 1, wherein the tile designation includes a compressed image included in the compressed image. An image decompression device, which decompresses based on information of a square tile specified by a means. 20. The image decompression device according to claim 19, further comprising an entropy decoder, an inverse quantizer, and an inverse two-dimensional discrete wavelet transformer. 21. A tile designating step of designating a rectangular tile with an arbitrary size for an image, an image dividing step of dividing an image using the designated rectangular tile, A bit plane decomposing step of decomposing into bit planes for each tile, a bit plane prioritizing step of prioritizing the decomposed bit planes in accordance with the encoding order, and a layer configuration forming a layer by the ranked bit planes And ranking the bit planes in the region near the tile boundary that keeps the quantization rate relatively low with respect to the bit planes in the region that is far from the tile boundary that keeps the quantization ratio relatively high. Layer so that it is included in the upper layer of the An image compression method characterized by differentiating. 22. A tile specifying step of specifying a rectangular tile of an arbitrary size for an image, an image dividing step of dividing an image using the specified rectangular tile, and A bit plane decomposing step of decomposing into bit planes for each tile, a bit plane prioritizing step of prioritizing the decomposed bit planes in accordance with the encoding order, and a layer configuration forming a layer by the ranked bit planes And a step of designating an area in which the quantization rate is kept low by selective area processing. 23. A tile designating step of designating a rectangular tile with an arbitrary size for an image, an image dividing step of dividing an image using the designated rectangular tile, A bit plane decomposing step of decomposing into bit planes for each tile, a bit plane prioritizing step of prioritizing the decomposed bit planes in accordance with the encoding order, and a layer configuration forming a layer by the ranked bit planes An image compression method comprising: storing a coefficient value before quantization of an area in which a quantization rate is suppressed to be low. 24. An image decompression method for decompressing a compressed image compressed by the image compression method according to any one of claims 21 to 23, wherein the compressed image is included in the compressed image. An image decompression method characterized in that decompression is performed based on information on a square tile specified by a means. 25. The image compression apparatus according to claim 1, wherein the image compression apparatus includes:
Alternatively, a program for causing a computer to function as the image decompression device according to claim 19 or 20 or each unit of the image decompression device. 26. The image compression apparatus according to claim 1 or each of the means of the image compression apparatus,
21. A computer-readable recording medium in which a program for causing a computer to function as the image decompression device according to claim 19 or 20 or each unit of the image decompression device is recorded.