JP2007295503A - Method and device for compressing image using method for encoding hierarchy - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To obtain a method and a device for compressing an image reproducing the image at a high compression rate without deteriorating an image quality by reducing the redundancy of a data list output by a hierarchical encoding method. <P>SOLUTION: An image information is encoded orthogonally, and the data list output from the method for encoding the hierarchy sorting a transfer coefficient into a significant transfer coefficient and a nonsignificant one regarding a quantizing-object bit plane by a tree structure is partitioned into blocks and block codes are generated. The generated block codes are replaced with other variable-length codes having a high encoding efficiency by using an entropy encoding, and the compression rate is improved. <P>COPYRIGHT: (C)2008,JPO&INPIT

Description

  The present invention relates to image information compression and a digital video signal quantizer, and more particularly to an image compression method and image compression apparatus using a hierarchical encoding method which is one of image information compression methods.

As a hierarchical encoding method, a set partitioning (SPIHT) method in a hierarchical tree structure is proposed by Non-Patent Document 1 described later.
In brief, in the SPIHT method, after discrete wavelet transform is performed on image information, transform coefficients located in the same place or in the same direction between frequency bands are associated by a tree structure. The root of the tree structure is located in the lowest frequency band, and every node has four children. In the SPIHT method, parent-child transform coefficients associated between different frequency bands depending on a tree structure are classified into significant transform coefficients and insignificant transform coefficients using a list in the quantization target bit plane. Whether or not the non-significant transform coefficient registered in the list is a significant transform coefficient is checked every time the bit plane to be quantized is changed. If it is a significant conversion coefficient, a polarity bit is output. Further, every time the quantization target bit plane is changed, the bit plane values of already significant transform coefficients registered in a list different from the insignificant transform coefficient group list are output in order from the most significant bit. Therefore, in the SPIHT method, encoding is performed in the order from the most significant bit plane to the least significant bit plane, so that an optimum image can be reproduced even if the encoding is finished at an arbitrary stage.

On the other hand, a non-patent document 2 described later also proposes a WQT hierarchical coding method.
In this hierarchical coding method of the WQT method, after discrete wavelet transform, the transform coefficients are correlated by a tree structure within each frequency band. In this hierarchical coding method, the SPIHT method is used to classify adjacent transform coefficients into significant transform coefficients and insignificant transform coefficients in the quantization target bit plane using a tree structure division and list by a quadtree. The encoding procedure of the insignificant transform coefficient group and the already significant transform coefficient group by the list is the same as that of the SPIHT method.

  A. Said and W.M. Pearlman, “A new fast and effective image codec based on setting in hierarchical trees”, IEEE Transactions Circuits and Systems Video Systems. 6, no. 3, pp. 243-250, June 1996   A. Munteanu, J.A. Cornelis, G.M. V. der Auwera, and P.M. Criste, “Wavelet image compression-The quadtree coding approach”, IEEE Transactions Information Technology in Biomedicine, vol. 3, no. 3, pp, 176-185, Sept. 1999

  However, in the conventional hierarchical encoding method described above, when the non-significant transform coefficients and the significant transform coefficients are classified from the non-significant transform coefficient group registered in the list, and the polarity bit is output if significant, There is a bias in the appearance probability of the output data series.

  Also, in the conventional hierarchical encoding method, when classifying significant transform coefficients and insignificant transform coefficient groups in the bit plane to be quantized based on the tree structure, the appearance probability of the data series obtained by encoding the significant transform coefficients Is biased.

  Further, when the bit plane values of already significant transform coefficient groups registered in the list are output in order from the most significant bit, the appearance probability of the output data series is also biased.

  The present invention has been made in view of such circumstances, and by reducing the redundancy of data sequences whose appearance probabilities are biased in the conventional hierarchical encoding method, the same quantized bit plane as in the conventional hierarchical encoding method is provided. The purpose of the present invention is to obtain an image compression method and an image compression apparatus using a hierarchical encoding method capable of reproducing an image having the same image quality as a conventional hierarchical encoding method even at the end of encoding with a high compression rate. Yes.

  In order to meet such an object, an image compression method using a hierarchical encoding method according to the present invention (the invention according to claim 1) is an image compression method for compressing image information using a hierarchical encoding method. After the image information is orthogonally transformed by discrete wavelet transform or discrete cosine transform, significant transformation is performed when classifying transform coefficients associated with a tree structure into significant transform coefficients and non-significant transform coefficient groups in the bit plane to be quantized. A block code is generated by dividing a data sequence obtained by coefficient coding into blocks, and the generated block code is replaced by another variable length code having high coding efficiency by entropy coding.

  The image compression method using the hierarchical encoding method according to the present invention (the invention described in claim 2) is the data compression method according to claim 1, wherein the data series obtained by encoding the insignificant transform coefficient group registered in the list is used. A block code is generated by dividing into blocks, and the block code generated by using entropy coding is replaced with another variable length code having high coding efficiency.

  The image compression method using the hierarchical encoding method according to the present invention (the invention described in claim 3) is the data compression method according to claim 1, wherein the data series obtained by encoding the significant transform coefficient group registered in the list is used. A block code is generated by dividing into blocks, and the block code generated by using entropy coding is replaced with another variable length code having high coding efficiency.

  An image compression method using the hierarchical coding method according to the present invention (the invention according to claim 4) is the variable compression code table according to any one of claims 1 to 3, or an adaptive entropy code. The block code is replaced with another variable-length code with high coding efficiency by conversion.

  The image compression method using the hierarchical coding method according to the present invention (the invention according to claim 5) is the average code length of the variable length code assigned to the block code according to any one of claims 1 to 4. However, a block code having a code length that approaches the lower limit of the average code length as much as possible is used.

  An image compression method using the hierarchical encoding method according to the present invention (invention according to claim 6) is the adaptive compression method according to any one of claims 1 to 5, wherein adaptive entropy encoding using a Markov information source model is used. The block code is replaced with another variable-length code having high coding efficiency.

  The image compression method using the hierarchical coding method according to the present invention (the invention described in claim 7) is the image compression method according to any one of claims 1 to 6, wherein the image information is obtained by discrete wavelet transform, discrete cosine transform, or the like. In order to further reduce the redundancy of the data sequence output by encoding transform coefficients using any of the image compression methods after orthogonal transform, adaptive arithmetic codes using Markov information source models are combined organically And

  An image compression apparatus using the hierarchical coding method according to the present invention (the invention according to claim 8) performs image compression using the image compression method according to any one of claims 1 to 7. It is comprised by these.

  That is, according to the present invention, a data sequence obtained by encoding an insignificant transform coefficient group registered in a list is divided into blocks, and a block code is generated. Then, the block code generated by using entropy coding is replaced with another variable length code having high coding efficiency to improve the compression rate.

  Also, according to the present invention, when classifying significant transform coefficients and insignificant transform coefficient groups in a quantization target bit plane based on a tree structure, a data sequence obtained by encoding significant transform coefficients is stored in one block. To generate a block code. Then, the block code generated by using entropy coding is replaced with another variable length code having high coding efficiency to improve the compression rate.

  Furthermore, according to the present invention, a block code is generated by dividing a data sequence obtained by encoding a significant transform coefficient group registered in the list into blocks of length m. The block code generated by using entropy coding is replaced with another variable length code having high coding efficiency to improve the compression rate.

  As described above, according to the image compression method and the image compression apparatus using the hierarchical encoding method according to the present invention, the redundancy of the data sequence output by the conventional hierarchical encoding method is reduced by entropy encoding. is there. Here, entropy coding is a compact code that has the shortest average code length. The compact code is an instantaneous code and can be uniquely decoded. In addition, when the length of a block code is increased as much as possible according to Shannon's first basic theorem, also known as the information source coding theorem, the average code length of the variable length code assigned to the block code becomes the lower limit of the average code length. Proven to be infinitely close.

  Therefore, the image compression method and the image compression apparatus using the hierarchical encoding method according to the present invention are similar to the conventional hierarchical encoding method even if the encoding is completed with the same quantization bit plane as the conventional hierarchical encoding method. There is an excellent effect that an image having the same image quality can be reproduced at a high compression rate.

  1 to 3 show an embodiment of an image compression method and an image compression apparatus using the hierarchical coding method according to the present invention.

  Here, D (i, j) represents a set of coordinates of all descendants of the node (i, j) of the tree. O (i, j) represents a set of coordinates of all children of the node (i, j). When node (i, j) is a leaf, O (i, j) is an empty set. L (i, j) is the coordinates of all descendants of the node (i, j) of the tree excluding all children as L (i, j) = D (i, j) −O (i, j). Represents a set. H represents a coordinate set of transformation coefficients of all tree roots. Note that (i, j) is the coordinates of the conversion coefficient.

  Further, the coordinates of transform coefficients that are not significant in the quantization target bit plane are registered in a list LIP (List of Insignificant Pixel). The coordinates of significant conversion coefficients are registered in a list LSP (List of Significant Pixel). A list LIS (List of Insignificant Set) is used to represent a set of insignificant transform coefficients using a tree and register the set.

ここで、Ｃ（ｉ，ｊ）は座標（ｉ，ｊ）における変換係数を表す。The number of bit planes to be quantized is obtained by the following equation 1.

Here, C (i, j) represents a conversion coefficient at coordinates (i, j).

An algorithm in which the image compression method according to the present invention is applied to the SPIHT method shown in FIG. 1 will be described in detail below.
That is, FIG. 1 shows a flowchart of an algorithm of an image compression method by the SPIHT method.

1. Initialization In the initialization in step S101, the value of the number n of bit planes to be quantized is output. Initialize the LSP to an empty set. Also, LIS is initialized to coordinates (i, j) ∈H of all tree roots, and LIP is initialized to coordinates (i, j) ∈H representing transformation coefficients of all tree roots.

2. Sorting path 2.1 The first element (i, j) of LIP is selected to start the sorting path. Until the last element of LIP is encoded, the next element (i, j) of LIP is selected and LIP encoding is repeated.

2.1.1 Output Sn (i, j) in step S102.
2.1.2 If significant Sn (i, j) = 1 in step S103, the polarity bit of the significant transform coefficient C (i, j) is output in step S104, and at the same time, the element (i, j) from LIP to LSP j) is moved.
2.1.3 In step S201, a data sequence obtained by encoding LIP elements is divided into blocks to generate block codes. In order to replace the generated block code with another variable length code having high coding efficiency, entropy coding is performed in step S202.

2.2 LIS encoding starts with the first element (i, j) of the LIS.
2.2.1 If element (i, j) is type A in step S105, Sn (D (i, j)) is output in step S106. If significant D (i, j) = 1 in step S107, Sn (k, l) is output in step S108. If significant Sn (k, l) = 1 in step S109, the polarity bit of the significant conversion coefficient C (i, j) is output in step S110, and at the same time, the element (k, l) is added to the LSP. On the other hand, if not significant such as Sn (k, l) = 0 in step S109, the element (k, l) is added to the LIP. The encoding is continued until the encoding of (k, l) εD (i, j) is completed in step S111.
After the encoding of the type A element (i, j), if L (i, j) ≠ φ in step S112, (i, j) is set as type B in step S113 and moved to the end of the LIS. If L (i, j) ≠ φ is not satisfied in step S112, (i, j) is deleted from the LIS in step S117.

2.2.2 In step S301, a block code is generated with a data sequence obtained by encoding type A element (i, j) as one block. In order to replace the generated block code with another variable length code having high coding efficiency, entropy coding is performed in step S302.

2.2.3 If element (i, j) is type B in step S105, Sn (L (i, j)) is output in step S114. If significant Sn (L (i, j)) = 1 in step S115, (k, l) ε0 (i, j) is added as type A to the end of the LIS in step S116, and at the same time in step S117. Delete (i, j) from the LIS. On the other hand, if it is not significant such as Sn (L (i, j)) = 0 in step S115, the next element (i, j) of LIS is selected until the last element of LIS is encoded. Repeat LIS encoding.

3. Refinement Path 3.1 Select the first element (i, j) of the LSP and start the refinement path. In step S118, the nth upper bit of the significant conversion coefficient C (i, j) is output. Until the last element of the LSP is encoded, the next element (i, j) of the LSP is selected and the LSP encoding is repeated.

3.2 In step S401, the data series obtained by the refinement pass is divided into blocks of length m to generate block codes. In order to replace the generated block code with a variable-length code with high coding efficiency, entropy coding is performed in step S402.

4). Updating the Quantized Bit Plane After the calculation of n = n−1 in step S119, if the bit plane to be quantized has not been reached, the procedure is repeated from the sorting path. If it has reached, the encoding is terminated.

An image compression method using the hierarchical coding method according to the present invention will be described in detail with reference to an algorithm applied to the WQT method shown in FIG.
FIG. 2 is a flowchart of the algorithm.

1. Initialization In the initialization in step S501, the value of the number n of bit planes to be quantized is output. Initialize list LIP and list LSP to an empty set. The list LIS is initialized to the tree root SQ (0, 0, t). SQ (0, 0, t) represents an area having root coordinates (0, 0) and size t × t.

2. LIP Encoding 2.1 Select the first element (i, j) of LIP and start the sorting path. Until the last element of LIP is encoded, the next element (i, j) of LIP is selected and LIP encoding is repeated.
2.1.1 If the transform coefficient is significant in step S502, a significant bit SGN symbol is output in step S503. Next, the polarity bit of the significant conversion coefficient C (i, j) is output in step S504.
2.1.2 Then move element (i, j) from LIP to LSP.
2.1.3 On the other hand, if the transform coefficient is insignificant in step S502, an insignificant bit NSG symbol is output in step S505.

2.2 In step S201, the data sequence obtained by encoding the LIP elements is divided into blocks to generate block codes. In order to replace the generated block code with another variable length code having high coding efficiency, entropy coding is performed in step S202.

3. Significant path 3.1 LIS encoding starts with the first element SQ (i, j, t) of the LIS. Select the next element (i, j) of the LIS and repeat the significant pass until the last element of the LIS is encoded.
3.1.1 In step S506, the presence or absence of a significant conversion coefficient is checked in a region where the upper left coordinate is (i, j) and the size is t × t. If not, an NSG symbol is output in step S507.

3.1.2 If there is a significant transformation coefficient, after outputting the SGN symbol in step S508, SQ (i, j, t) is converted into a quadtree SQ (i, j, t / 2) in step S509, SQ (i + t / 2, j, t / 2),
SQ (i, j + t / 2, t / 2), SQ (i + t / 2, j + t / 2, t / 2)
It expresses. If the region represented by the quadtree is composed of two or more transform coefficients in step S510, the quadtree is added as an element at the end of the LIS in step S511, and at the same time, from the LIS to SQ (i , J, t).

3.1.3 On the other hand, if the region represented by the quadtree in step S510 is SQ (i, j, 1) composed of one transform coefficient, if the transform coefficient is significant in step S513, step S513 is performed. In S514, the significant bit SGN symbol is output. Next, after outputting the polarity bit of the significant conversion coefficient C (i, j) in step S515, the coordinates (i, j) are moved to the LSP. On the other hand, if the transform coefficient is insignificant in step S513, the insignificant bit NSG symbol is output in step S516, and then the coordinates (i, j) are moved to LIP. The encoding procedure of SQ (i, j, 1) is repeated until the entire region composed of one transform coefficient is encoded.

3.2 In step S301, after encoding of all SQ (i, j, 1) is completed, a block code is generated with the obtained data series as one block. In order to replace the generated block code with another variable length code having high coding efficiency, entropy coding is performed in step S302.

4). Refinement path The first element (i, j) of the LSP is selected to start the refinement path.
4.1 Output the nth upper bit of the significant conversion coefficient C (i, j) in step S517. Until the last element of the LSP is encoded, the next element (i, j) of the LSP is selected and the LSP encoding is repeated.

4.2 In step S401, the data series obtained by the refinement pass is divided into blocks of length m to generate block codes. In order to replace the generated block code with a variable-length code with high coding efficiency, entropy coding is performed in step S402.

5). Updating the Quantized Bit Plane After the calculation of n = n−1 in step S518, if the bit plane to be quantized has not been reached, the procedure is repeated from LIP coding. If it has reached, the encoding is terminated.

FIG. 3 is a block diagram showing the configuration of the image compression apparatus according to the embodiment of the present invention.
The configuration of an image compression apparatus that implements the above-described hierarchical encoding method will be described below with reference to FIG.

  In FIG. 3, code 11 is an image information conversion unit in the encoder, 12 is an orthogonal transform unit such as DCT and DWT, 13 is a hierarchical coding unit, and 14 is an adaptive arithmetic coding unit using a Markov information source model. The encoder operates as follows.

  First, image information that is an original image is input to the image information conversion unit 11. When an image is composed of signals having no polarity, the signal value is level shifted to a range centered on zero. Next, when the image information is composed of color signals, the color signals are converted into luminance / color difference components by converting the image information. This conversion of image information is effective in reducing the correlation between color signals.

  The orthogonal transform unit 12 decomposes image information composed of luminance and color difference components into frequency bands by orthogonal transform such as discrete cosine transform (DCT) and discrete wavelet transform (DWT).

Next, the above-described hierarchical encoding unit 13 encodes the transform coefficient for each frequency band from the most significant bit plane. The number of bit planes to be quantized is changed for each frequency band, and the compression rate is increased without significantly degrading the image quality.
For the entropy coding in the hierarchical coding method described above, adaptive entropy coding using a Markov information source model or entropy coding using a variable length code table 15 is used.

  The adaptive arithmetic coding unit 16 to which the Markov information source model is applied further reduces the redundancy of the data series output from the hierarchical coding unit 13. The data series output from the encoder is stored or transmitted on the path 16 and sent to the decoder.

  The decoder is configured as follows. 3, reference numeral 17 denotes an adaptive arithmetic decoding unit using a Markov information source model in the decoder, 18 denotes a hierarchical decoding unit, 19 denotes an orthogonal inverse transformation unit such as IDCT and IDWT, and 20 denotes an image information inverse transformation unit. It is.

In this decoder, the data sequence via the adaptive arithmetic decoding unit 17 to which the Markov information source model is applied is converted into transform coefficients for each frequency band by the hierarchical decoding unit 18.
At this time, adaptive entropy decoding using the Markov information source model or entropy decoding using the variable length code table 15 is used. Since decoding is performed in the order from the most significant bit plane to the least significant bit plane, an optimum image can be reproduced even if hierarchical decoding is completed with an arbitrary bit plane.

  That is, the transform coefficient decomposed into frequency bands by the orthogonal inverse transform unit 19 such as inverse discrete cosine transform (IDCT: Inverse Discrete Wavelet Transform) or inverse discrete wavelet transform (IDWT: Inverse Discrete Wavelet Transform) is converted from the luminance and chrominance components. Is converted into image information.

  Then, the image information inverse transform unit 20 of this decoder can perform color inverse transform and level inverse shift to obtain a reproduced image.

  According to such an image compression method according to the present invention, the data series obtained by encoding the insignificant transform coefficient group registered in the list is divided into blocks, a block code is generated, and the generated block code is By replacing with another variable-length code having high coding efficiency, the compression rate can be improved.

  Also, according to the present invention, when classifying significant transform coefficients and insignificant transform coefficient groups in a quantization target bit plane based on a tree structure, a data sequence obtained by encoding significant transform coefficients is stored in one block. As a block code is generated, and the generated block code is replaced with another variable length code having high encoding efficiency, the compression rate can be improved.

  Furthermore, according to the present invention, the data series obtained by encoding the significant coefficient group registered in the list is divided into blocks of length m, a block code is generated, and the generated block code is encoded. The compression rate can be improved by replacing with another variable length code having high conversion efficiency.

  According to the image compression method and the image compression apparatus using such a hierarchical encoding method, the redundancy of the data sequence output by the conventional hierarchical encoding method can be reduced by entropy encoding. Even if the encoding is completed with the same quantization bit plane as the encoding method, an image having the same image quality as that of the conventional hierarchical encoding method can be reproduced with a high compression rate.

  Note that the present invention is not limited to the structure described in the above-described embodiment, and the shape and structure of each part constituting the image compression method and the image compression apparatus using the hierarchical coding method can be appropriately modified and changed. Needless to say.

  In particular, in the above-described embodiment, the case where image compression is performed using the encoding method based on the SPIHT method and the WQT method has been described. However, the present invention is not limited to this, and various conventionally known encoding methods are used. The present invention can be applied to an image compression method based on the method and exhibit an effect.

  It is a flowchart for demonstrating the image compression method using the hierarchical encoding method which concerns on this invention, Comprising: The encoding method of this invention based on a SPIHT system.   It is a flowchart for demonstrating the image compression method using the hierarchical encoding method which concerns on this invention, Comprising: The encoding method of this invention based on a WQT system.   It is a block diagram which shows the structure of the image compression apparatus in embodiment of this invention.

Explanation of symbols

  DESCRIPTION OF SYMBOLS 11 ... Image information conversion part, 12 ... Orthogonal transformation part, 13 ... Hierarchical coding part, 14 ... Adaptive arithmetic coding part, 15 ... Variable length code table, 16 ... Transmission or accumulation (path | route), 17 ... Adaptive arithmetic decoding part, 18: Hierarchical decoding unit, 19 ... Orthogonal inverse transform unit, 20 ... Image information inverse transform unit.

Claims (8)

An image compression method for compressing image information using a hierarchical encoding method,
After orthogonally transforming image information using discrete wavelet transform or discrete cosine transform, significant transform coefficients are used when classifying transform coefficients associated with a tree structure into significant transform coefficients and insignificant transform coefficient groups in the quantization target bitplane. Hierarchical coding characterized in that a block code is generated by dividing the data sequence obtained by encoding into a block, and the generated block code is replaced with another variable-length code with high encoding efficiency by entropy coding An image compression method using the method. The image compression method using the hierarchical encoding method according to claim 1,
The data series obtained by encoding the nonsignificant transform coefficient group registered in the list is divided into blocks, block codes are generated, and block codes generated using entropy encoding are separated from those with high encoding efficiency. An image compression method using a hierarchical encoding method, characterized in that it is replaced with a variable length code. The image compression method using the hierarchical encoding method according to claim 1,
The data sequence obtained by encoding the significant transform coefficient group registered in the list is divided into blocks, block codes are generated, and block codes generated using entropy encoding are separated from those with high encoding efficiency. An image compression method using a hierarchical encoding method, characterized in that it is replaced with a variable length code. The image compression method using the hierarchical encoding method according to any one of claims 1 to 3,
An image compression method using a hierarchical coding method, wherein a block code is replaced with another variable length code having high coding efficiency by using a variable length code table or by adaptive entropy coding. An image compression method using the hierarchical encoding method according to any one of claims 1 to 4,
An image compression method using a hierarchical encoding method, wherein a block code having a code length such that an average code length of a variable-length code assigned to a block code approaches the lower limit of the average code length as much as possible is used. In the image compression method using the hierarchical encoding method according to any one of claims 1 to 5,
An image compression method using a hierarchical coding method, wherein a block code is replaced with another variable-length code having high coding efficiency by adaptive entropy coding using a Markov information source model. An image compression method using the hierarchical coding method according to any one of claims 1 to 6,
In order to further reduce the redundancy of the output data sequence after encoding the transform coefficients using one of the image compression methods after orthogonal transformation of the image information by discrete wavelet transform or discrete cosine transform, a Markov information source model is used. An image compression method using a hierarchical coding method characterized by organically combining adaptive arithmetic codes.   An image compression apparatus using a hierarchical encoding method, wherein the image compression method is configured to perform image compression using the image compression method according to any one of claims 1 to 7.

Images