JPH0563989A - Encoder of image signal - Google Patents

Abstract

PURPOSE:To improve the reproduction picture quality by performing the adaptive coefficient selection to the coefficient distribution different for each picture element block in the transformed coding system to perform the adaptation processing. CONSTITUTION:By a blocking device 101, a rectangular region in an image is segmented as a picture element block from a digital image signal, an orthogonal transformation is performed by a transformation unit 103 and the transformation coefficients are obtained. An activity calculator 211 calculates the precision of an image in the block and a waveform analyzer 205 analyzes the waveform of the image. From the obtained precision and waveform information, an optimum mask is selected from a mask dictionary 209 by a mask selector 208. Based on the mask, the transformation coefficients are selected by a coefficient selector 105. The transformation coefficients after selection are quantized by a quantizer 107, coded by an encoder 109 and subsequently, outputted to a transmission line 111.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an image signal coding apparatus.

[0002]

2. Description of the Related Art When gradation image information is handled, the amount of information becomes very large if the image information is digitized as it is. Therefore, the amount of information is generally compressed by encoding the image information.

There are various methods for encoding image information, but as a typical encoding method for gradation images, Hideo Hashimoto: Introduction to image information compression "Image encoding algorithm II-conversion encoding-", television John Society Journal, Vol.
43, No. 10 (1989), pp. 1145-11
A transform coding method such as that disclosed in No. 52 is known.

The structure of this transform coding method will be described with reference to the basic block diagram shown in FIG.

Reference numeral 101 is a blocker that cuts out a rectangular area in the image as a pixel block 102 from the digital image signal 100, 103 is a converter that performs orthogonal transform on the pixel block 102 and outputs it as a transform coefficient 104, and 105 is a transform coefficient 104 Select a specific coefficient and select conversion coefficient 106
, A quantizer 107 for quantizing the selected transform coefficient 106 and outputting a quantized coefficient 108,
Reference numeral 9 is an encoder that encodes the quantized coefficient 108 and outputs the encoded data 110 to the transmission path 111.

Next, the operation will be described. In the example shown in FIG. 11, the encoding process includes a conversion process, an information reduction process, and a code allocation process.

In the conversion process, in the case of an image signal, a two-dimensional orthogonal conversion utilizing the correlation in the horizontal and vertical directions is used. The blocker 101 forms a pixel block 102 composed of M pixels and N pixels in the horizontal and vertical directions, respectively, and the converter 103 performs one-dimensional orthogonal transformation independently in the horizontal and vertical directions. In the converter 103,
The linear conversion shown in equation (1) is performed.

Y = A _N XA _M ^T (1) Here, X is a pixel block 102 of N rows and M columns, respectively.
Y is a conversion coefficient 104, and A _N and A _M are N
Next, it is an M-th order orthogonal transformation matrix.

There are various types of orthogonal transform, but in terms of coding efficiency, discrete cosine transform (hereinafter referred to as DCT).
Is generally used. The transformation of the two-dimensional DCT is given by equation (2), and the inverse transformation is equation (3).

[0010]

[Equation 1]

Here,

[0012]

[Equation 2]

Further, X (j, k) is the pixel block 10
2 represents each element, and j and k represent the position of the element. Y
(U, v) represents each element of the transform coefficient 105, and u, v
Indicates the position of the element.

The information reduction processing is performed by the coefficient selector 105 and the quantizer 107. In the coefficient selector 105,
A coefficient is selected based on the variance of the conversion coefficient 104 to obtain the selected conversion coefficient 106.

In such a transform coding method, a method of selecting a coefficient larger than the threshold value by comparing the variance of the coefficient with a constant threshold value and setting the coefficient equal to or less than the threshold value to 0 to improve the compression efficiency is William K. ． Pratt:
"Digital Image Processin
g ", Wiley-Interscience, p
p. 678-699. The threshold value in this case can be determined from the statistics of the transform coefficients of many images. Also, a method of determining from the statistics of the conversion coefficient for each image has been proposed. However, when the distribution of the conversion coefficient is different from these statistics, image quality deterioration may occur.

In the quantizer 107, the selective transform coefficient 106
Is quantized to obtain a quantized coefficient 108.

The code allocation process is performed by the encoder 109.
In, the coded data 110 is constructed by assigning a code word to the quantization coefficient 108, and is output to the transmission path 111.

By each of the above processes, the image information can be coded by the transform coding method.

However, in the above-mentioned method, since the selection of the coefficient is uniformly determined for all the pixel blocks, there is a problem that it cannot be applied to the variation of the local property of the image.

In order to solve this problem, W. H. Che
n. et al: “Adaptive Coding
of Monochrome and Color I
mages ”, IEEE Transactions
on Communications. , VOL. CO
M-25, No. 11, pp. 1285-1292 (N
ov. As disclosed in (1977), a method of performing adaptation for each block has been proposed. This technique
The image blocks are classified in advance into four classes according to the magnitude of the AC power in the blocks, and the coefficient selection criterion (bit allocation) is determined from the coefficient variance obtained for each class.

In this method, the classification is performed based only on the magnitude of the AC power in the block. Therefore, even if the AC power in the block is the same, the coefficient distribution may differ due to the directionality such as edges. is there. However, in this method, since the coefficient at the same position is simply selected within the same class regardless of the difference in coefficient distribution, the most suitable coefficient for expressing the input image cannot be selected, It was not possible to obtain a sufficiently high quality image.

In view of the above problems, Kato, Takekawa, and Okubo: "Adaptive Orthogonal Transform Coding Scheme Utilizing Classification", IEICE Transactions (B), Vol. J71
-B, No. 1, pp1-9, January 1988, proposed a block classification method using a vector quantization method. In this method, by classifying considering the magnitude and bias of the AC power in the block,
The W. H. Chen. It is reported that the image quality and performance are improved as compared with the method described in the et al document.

In the adaptive orthogonal transform coding method using the above classification, the coefficients used as vectors are limited to those in the low frequency band where power is concentrated in order to reduce the enormous amount of calculation required for vector quantization. There is. For this reason, it cannot be said that the case where the coefficient is distributed up to a high frequency band, such as a pixel block including a steep edge, is considered. Further, even if the dimension of the vector is expanded and the vector quantization including even the high-frequency coefficient is performed, the high-frequency coefficient rarely follows a specific distribution, so that the effect of classifying cannot be expected. The present invention has been devised to solve the above problems, and an object of the present invention is to improve reproduced image quality by performing adaptive coefficient selection for different coefficient distributions for each pixel block. To do.

[0024]

In order to achieve the above-mentioned object, an image signal coding apparatus of the present invention codes an image signal for coding a transform coefficient after dividing the image signal into blocks and performing orthogonal transformation. The image signal is N × M
A means for dividing a pixel block into rectangular blocks (N and M are positive integers), a means for performing a two-dimensional orthogonal transformation on the pixel block, and a waveform in the pixel block are compared with a plurality of representative patterns, and are most similar to each other. The waveform analysis means for outputting the identification information of the representative pattern, the storage means for registering a plurality of mask information indicating the significance / insignificance of each transform coefficient obtained from the two-dimensional orthogonal transformation means, and the waveform analysis means. Based on the identification information, a mask selection unit that selects mask information from the storage unit and the selected mask information are used to determine whether the conversion coefficient is significant or insignificant, and the conversion coefficient determined to be insignificant is set to 0. It is characterized by comprising coefficient selecting means and encoding means for encoding the selected transform coefficient.

[0025]

According to the present invention, when the image signal is encoded, the waveform in the pixel block is compared with a plurality of representative patterns and the identification information of the most similar representative pattern is output. A plurality of mask information for replacing the conversion coefficient with 0 is prepared in advance corresponding to each representative pattern,
By masking the transform coefficient based on the identification information, the coefficient at the optimum position according to the content of the image is encoded.

[0026]

1 is a block diagram showing an embodiment of an encoding apparatus according to the present invention. The parts corresponding to those of the conventional example shown in FIG. 11 are designated by the same reference numerals.

101 is a blocker, 103 is a converter,
Reference numeral 105 is a coefficient selector, 107 is a quantizer, 109 is an encoder, and 111 is a transmission line. Further, 200 is an average value calculator that calculates an average value 201 from the pixel block 102, 202 is an average value separator that subtracts the average value 201 from each pixel value of the pixel block 102 and outputs an average value separation signal 203, Reference numeral 204 is a dispersion calculator that calculates the intra-block variance from the average value separation signal 203 and outputs it as activity 206. Reference numeral 205 is a waveform that analyzes the signal waveform in the block from the average value separation signal 203 and outputs it as waveform information 207. The analyzer 208 is a mask selector that selects the mask information 210 from the mask dictionary 209 based on the activity 206 and the waveform information 207. The selected mask information 210 is input to the coefficient selector 105. The average value calculator 200 and the average value separator 2
02 and the distributed calculator 204, the activity calculator 21
1 is configured.

In this embodiment, an adaptive process is performed in addition to the three processes of conversion, information reduction, and code allocation performed in the conventional example shown in FIG. This adaptive processing operates in parallel with the conversion processing, and realizes adaptive information reduction suited to the characteristics of each pixel block in the subsequent information reduction processing.

Since the three processes of conversion, information reduction, and code allocation are the same as those of the conventional example shown in FIG. 11, description thereof will be omitted, and only the adaptive process will be described.

The adaptive processing consists of activity calculation and waveform analysis.

The activity is an index representing the definition of the image, and in the following, the variance within the pixel block will be used. The activity σ is calculated by the following equation.

[0032]

[Equation 3]

Here, x _i is the pixel value within the block, m is the average value of the pixel values within the block, and N is the total number of pixels within the block.

The activity calculation will be described with reference to FIG. Pixel block 1 in average value calculator 200
The average value 201 of 02 (m in equation (5)) is calculated.
The average value separator 202 generates an average value separation signal 203 by subtracting the value of the average value 201 from each pixel value of the pixel block 102. The variance calculator 204 calculates the sum of absolute values of the average value separated signal 203 according to the equation (5), then divides the sum of absolute values by the total number of pixels in the block to calculate the activity 206, and outputs the result. To do.

Next, the waveform analysis will be described with reference to FIG. Pattern matching is used as a method for determining the similarity of waveforms. This is because a plurality of basic waveforms y1, y2, ···, yi, ··, yk prepared in the waveform dictionary and the input image block are regarded as vectors, and the shortest distance waveform is selected by judging the distance between them. Then, the identifier i of the selected waveform is output.

As a method of analyzing the waveform of such an image signal, Japanese Patent Application No. 3-202129 filed by the present applicant is proposed.
There is a method proposed in the specification. Note that the waveform analysis is performed in this manner. In order to adaptively encode the image signal, the waveform of the image signal is analyzed before encoding, and the encoding parameter is controlled using the analysis result. Is effective.

The configuration of the waveform analyzer 205 will be described with reference to FIG. Reference numeral 203 is an average value separation signal, 310 is a waveform comparator, and 311 is a waveform dictionary in which a plurality of basic waveforms are registered. Reference numeral 207 is waveform information for identifying the waveform registered in the waveform dictionary 311.

The operation of the waveform analyzer 205 will be described below. In the waveform comparator 310, the input average value separation signal 203 is subjected to distance determination with respect to a plurality of waveforms registered in the waveform dictionary 311. An identifier for identifying each of the plurality of waveforms registered in the waveform dictionary 311 is given. The identification information of the basic waveform determined to be the closest as a result of the distance determination is output as waveform information 207.

In FIG. 4, (A), (B),
The distribution of pixel values in each block is horizontal gradation, vertical gradation, and diagonal edge, respectively, and the distribution of pixel values in an 8 × 8 pixel block is shown in (C), (a), (b), (c) of FIG. ) Indicates a coefficient matrix when DCT is performed on each pixel block. As can be seen from the figure, a specific coefficient distribution corresponds to the distribution of pixel values in the input pixel block. This indicates that the distribution form of the coefficients can be predicted from the signal waveform of the input pixel block.

For example, the waveform of the pixel block is shown in FIG.
If it has a horizontal gradation shown in (A), as shown in (a) of the figure, the coefficient to be coded after conversion may be only that in the first row of the coefficient matrix.
As long as it has the vertical gradation shown in FIG. 7B, the coefficient to be coded after conversion may be only the first column of the coefficient matrix as shown in FIG. On the other hand, if the pixel block includes diagonal edges as shown in FIG. 6C, it is necessary to set a wide range of coefficients near the diagonal line of the coefficient matrix as the encoding target as shown in FIG. is there.

As described above, by controlling the selection of the coefficient in accordance with the predicted distribution form of the coefficient, it is possible to improve the image quality deterioration near the edge, which is a problem in the conventional transform coding method. That is, since the distribution state of the coefficients can be predicted from the result of the waveform analysis, the coefficient can be efficiently selected based on the prediction result.

Similarly, the activity indicating the fineness in the pixel block can also be used as an index for determining the selection range of the coefficient.

As shown in FIG. 5, a block with a small activity (see FIG. 7B) is a block with a flat pattern and has a gradual pixel distribution, and a block with a large activity (FIG. 5A). (See) is a fine pixel distribution including edges and the like, which is a block with complicated patterns. In a pixel block with small activity, the correlation between pixels is high, and power is concentrated on low-frequency coefficients after conversion, so these may be selected. Further, in a pixel block having a large amount of activity, the coefficient power is widely dispersed in the coefficient matrix due to the influence of edges and the like, so it is necessary to set a wide selection range of coding coefficients.

The method of selecting the coding coefficient by the mask will be described below.

FIG. 6 is a diagram for explaining the contents of the mask dictionary 209 in FIG. The mask information registered in the mask dictionary is selected from the previously determined waveform information and activity for each block. The index is determined from the activity by appropriate threshold processing. For example, when the activity is σ, thresholds T ₀ , T ₁ ,
..., T _N-1 and indexes R ₀ , R ₁ , ..., R
The relationship of _N can be defined as follows. N is the number of thresholds to be set, which allows the activity to
It is classified into the (N + 1) class. The threshold may be set for each waveform information.

The sigma <When a T _0, when the index when an _{R 0 T 0 ≦ σ <T} 1, the index is _{R 1 · · · · · ·} T N-1 ≦ σ, index R _N The coefficient selecting operation in the coefficient selector 105 will be described with reference to FIG. The mask information 210 is a matrix having the same size as the transform coefficient 104, and each element is “1”.
Alternatively, it is composed of binary information of "0". The mask information 210 shown in FIG. 7 corresponds to the mask information in which the waveform information in FIG. 6 is “1” and the index is “1”. The information of "1" and "0" forming each element is the coefficient selector 10
This is the instruction information for selecting the coefficient in 5. The unselected coefficients are forced to zero. The selection result is output as the selection conversion coefficient 106. In addition, in the conversion coefficient 104 shown in FIG. 7, a circle indicates a non-zero coefficient, and a cross indicates a zero coefficient.

The information reduction processing in the quantizer 107 and the code allocation processing in the encoder 109 are the same as in the prior art.

The mask information 210 registered in the mask dictionary 209 needs to be determined prior to the encoding operation, and the number thereof depends on the number of pixel block states classified by the waveform analysis and the number of activity classes. equal.

Each mask information is determined according to a coefficient distribution peculiar to each state of the pixel block. The procedure will be described below.

First, the state of the pixel block cut out from the image is determined based on waveform analysis and activity threshold processing, and the conversion coefficient for each pixel block is classified for each state.

Subsequently, a representative coefficient distribution characteristic is determined for each set of classified conversion coefficients. As the distribution characteristic, a coefficient distribution matrix obtained by cumulatively adding absolute values of coefficients for each position in the conversion coefficient matrix can be used.

Hereinafter, according to the flow chart shown in FIG.
A procedure for determining mask information from the coefficient distribution matrix will be described.

The coefficient distribution matrix for each set is sorted in descending order of the dispersion value, and each pair of the dispersion value and the coefficient position is stored in a table format as shown in FIG. 9 (step 101). In the case of the 8th-order two-dimensional orthogonal transformation, the position coefficient is in the range of (0,0) to (7,7), but since (0,0) is a DC component, this is excluded and all AC coefficients are excluded. The same number as 63 storage areas are prepared. The contents of the storage area can be referenced by a pointer.

Prior to the start of processing, the pointer is set to the top of the table, ie the maximum variance value in the variance matrix. Further, "0" is set for all the elements in the mask information (step 102).

In the loop in the flowchart shown in FIG. 8, the operation of setting "1" in the mask information corresponding to the coefficient position is performed in order from the coefficient with the largest variance (steps 105 and 106). At the same time, the cumulative addition of the variance values is performed (step 103), and the process is terminated when the ratio of the variance to the total sum exceeds a predetermined threshold value (step 10).
4). Further, the predetermined threshold can be set individually for waveform analysis and activity.

FIG. 10 shows an example of the distribution of DCT coefficient power concentration ratios and mask information created with a power storage ratio of 80% as the termination condition. (A), (B), and (C) of the figure show the power concentration rates of the respective coefficients corresponding to different image blocks, and the underlined portions indicate the coefficient positions that satisfy the power conservation rate of 80%. is there. In addition, FIG.
(B) and (c) are mask information corresponding to (A), (B), and (C) in the same figure, and the mask information of the coefficient position underlined is "1". However, the mask information at the (0,0) position indicating the DC component is always "1".

By registering the mask information thus created in the mask dictionary, it is possible to select a mask for optimum coefficient selection according to the waveform analysis result and the activity calculation result at the time of encoding.

[0058]

According to the present invention, since the conversion coefficient is adaptively selected according to the local property of the image, it is possible to suppress the deterioration of the image quality due to the presence of edges and the like.

[Brief description of drawings]

FIG. 1 is a configuration diagram showing an embodiment of an encoding device of the present invention.

FIG. 2 is an explanatory diagram of waveform analysis.

FIG. 3 is a configuration diagram of a waveform analyzer.

FIG. 4 is an explanatory diagram showing an example of a distribution of pixel values within a block and a distribution of DCT coefficients.

FIG. 5 is an explanatory diagram showing a relationship between pixel blocks and activities.

FIG. 6 is an explanatory diagram of contents of a mask dictionary.

FIG. 7 is an explanatory diagram showing a coefficient selecting operation according to the present invention.

FIG. 8 is a flowchart showing a procedure for creating mask information according to the present invention.

FIG. 9 is an explanatory diagram showing a result of sorting coefficient variances.

FIG. 10 is an explanatory diagram showing an example of mask information determined from coefficient distribution characteristics.

FIG. 11 is a configuration diagram showing an example of a conventional encoding device.

[Explanation of symbols]

100: digital image signal, 101: blocker, 1
02: pixel block, 103: converter, 104: transform coefficient, 105: coefficient selector, 106: selected transform coefficient, 10
7: Quantizer, 108: Quantization coefficient, 109: Encoder, 110: Code data, 111: Transmission line, 200: Average value calculator, 201: Average value, 202: Average value separator,
203: average value separation signal, 204: variance calculator, 20
5: waveform analyzer, 206: activity, 207: waveform information, 208: mask selector, 209: mask dictionary,
210: mask information, 211: activity calculator,
310: Waveform comparator, 311: Waveform dictionary

 ─────────────────────────────────────────────────── ─── Continuation of the front page (72) Inventor Setsu Kunitake 2274 Hongo, Ebina City, Kanagawa Prefecture Fuji Zerox Co., Ltd.

Claims (3)

[Claims] 1. An image signal coding apparatus for coding a transform coefficient after blocking an image signal and performing orthogonal transformation, wherein the image signal is a rectangular region of N × M (N and M are positive integers). A means for dividing the pixel block into two-dimensional orthogonal transforms, and a waveform analysis for comparing the waveform in the pixel block with a plurality of representative patterns and outputting identification information of the most similar representative pattern. Means, storage means for registering a plurality of mask information indicating significance / insignificance of each transform coefficient obtained from the two-dimensional orthogonal transformation means, and a mask from the storage means based on the identification information from the waveform analysis means. Mask selecting means for selecting information, and coefficient selecting means for judging whether the conversion coefficient is significant or insignificant using the selected mask information and setting the conversion coefficient determined to be insignificant to 0 Encoding apparatus of the image signal, characterized in that a coding means for encoding the transform coefficients after the selection. 2. The image signal coding apparatus according to claim 1, wherein the waveform analysis means compares the waveform in the pixel block with the plurality of representative patterns by pattern matching. 3. The method according to claim 1, further comprising means for calculating an activity in the pixel block, wherein the mask selecting means selects mask information from the storage means using the identification information and the activity. 1. An image signal encoding device as described in 1.