JPH082083B2 - Multi-level image block coding device - Google Patents

Description

DETAILED DESCRIPTION [Outline] The present invention divides a multi-valued image into blocks composed of a plurality of pixels and compares the difference between the maximum gradation value and the minimum gradation value in the block with a threshold value. In a coding method in which a representative grayscale value in a block is obtained and a multi-valued image is coded for each block, by setting the threshold according to the spatial distribution of grayscale values in the block, the code of noise It is possible to restore the edge part with a small gradation difference while reducing the amount of conversion.

[Industrial applications]

The present invention relates to a block coding method for a multi-valued image, and more particularly to a block coding method for a multi-valued image capable of restoring an edge portion having a small gradation difference while reducing a noise coding amount.

In order to store image data whose information amount is orders of magnitude larger than numerical data, especially halftone images and color image data, or to transmit at high speed and high quality, the gradation value for each pixel must be It needs to be encoded with high efficiency.

[Conventional technology]

Conventionally, as a means for solving the above problems, a block coding method has already been proposed as shown below.

(1) IEICE Transactions, January 1987, Vol.J70-B, No.1 “Differential adaptive block coding of grayscale images” (2) IEICE research paper 86-02-05 “Grayscale images” (3) IEICE General Conference 1988 Proposal 1359 "Color image variable block size encoding" It is divided into blocks of pixels, and the image information in this block is divided into a few gradations (gradation components).
And information (resolution component) indicating how the above-mentioned representative gradation is arranged in the block, and the encoding is performed. Above (1), (2),
The method (3) is different in the method of giving the number of representative gradations representing a block and the method of obtaining the representative gradation value.

In the following, the image signal has 256 gradations (8 bits / pixel) and 4 ×
The case where a block is set to 4 pixels will be described.

Method (1) In this method, an image is divided into blocks each having a size of m × m pixels, and a value between the maximum gradation value and the minimum gradation value in the block is 2 ^n.
It linearly quantizes the level and transmits it in a bit plane format.

The image information is characterized by being divided into three components of a reference level, a difference value, and a resolution component (bit plane information).

Explaining the case of m = 4 and n = 2, the maximum value L _max and the minimum value L _min of the gradation value of each block are equally divided as shown in FIG.

L _A , L _D , P _i and Q _j are given by the following equations.

Reference level: L _A = (L _max + L _min ) / 2… Difference value: L _D = (L _max −L _min / 2…) In a grayscale image, a block having a smaller gradation change can be approximated with a smaller number of gradations.

Therefore, as shown in FIG. 7, L _D is compared with threshold values T ₁ and T _2, and ₁ , ₂ , 4 are determined according to the magnitude of gradation change in the block.
Approximately expressed by gradation.

Since a maximum of 4 gradations (gradation components) is used, a maximum of 2 bits of information (resolution component) is required for each pixel in order to specify gradation within a block. The 2-bit resolution components Φ ₁ and Φ ₂ are encoded in the bit plane format.

Method (2) This method divides an image into blocks of m × m pixels, and then 1, 2,
3 ... and the number of gradations are given, nonlinear quantization is performed, and this is transmitted in the bit plane format.

It is similar to determine the number of display gradations in the block by comparing with the thresholds T ₁ , T _2, ..., However, the representative gradation and the arrangement of the representative gradation are determined by using, for example, a K-means algorithm. It is characterized in that the representative gradations are determined and the representative gradations are set at non-uniform intervals.

FIG. 8 shows an example of expressing four gradations Q ₁ , Q ₂ , Q ₃ , and Q ₄ .

Method (3) In this method, after dividing the image into blocks of m × m pixels, the gradation X _ij in the block is divided into three modes according to the following equations, and representative gradations are obtained and approximated.

One gradation expression: MAX {X _ij } -MIN {X _ij } ≦ T ₁ 1 ≦ i, j ≦ m ... Multi-gradation expression: under conditions other than the above and / or conditions where MAX {•} and MIN {•} represent maximum and minimum values, Represents a gradation higher than the average value and a gradation lower than the average value, respectively.

As a representative gradation, an average value is used when expressing 1 gradation, and an average value is used when expressing 2 gradations. In case of multi-gradation expression, 1/2 sub-sampling is applied, and then 3-bit nonlinear quantization (DPCM) is applied to the prediction error signal.

T ₁ and T ₂ are threshold values. In this method, as shown in FIG. 9 as T ₁ , a variable threshold value T ₁ is used according to the block average value, and when blocks of the same gradation number are adjacent, Only when the areas of those blocks are included in the specified shape, the areas are integrated and the area where multiple blocks are connected is
The feature is that it is approximated with a small number of gradations.

In addition to the above method, a preprocessing method has been proposed in the following document in order to improve the compression rate when block-coding a halftone dot picture having characteristics such as a noisy image.

・ IEICE Transactions February 1987 Vol.No.J70-B, No.2 "Method for Discriminating Halftone Photographs" In this method, it is possible to discriminate halftone dot areas where black and white change in a short cycle. The procedure is as follows.

 FIG. 10 shows the flow of identification processing.

(1) The image signal is divided into blocks of N × N images.

(2) Obtain the maximum level MAXL and the minimum level MINL of the block image signal.

(3) The following judgment is made by comparing with a predetermined reference value P.

If MAXL-MINL <P, it is determined to be a non-halftone dot block.

If MAXL-MINL ≧ P, the following processing (4) is performed.

(4) Each pixel is binarized by the average value of the image signal in the block.

(5) Find the number of 0/1 changes K _H that occur between consecutive pixels in the block in the main scanning direction.

Similarly, in the sub-scanning direction, the number of 0/1 changes K _V
Ask for.

(6) The following judgment is made by comparing with a predetermined reference value T.

If K _H ≧ T and K _V ≧ T, it is determined as a halftone dot block.

If K _H <T or K _V <T, it is determined as a non-halftone dot block.

Then, in order to remove the halftone dot structure, the block determined as the halftone dot area is subjected to a smoothing filter of a mask of 3 × 3 pixels.

Then, the block for which the smoothing process has been performed is coded.

[Problems to be Solved by the Invention]

In the block encoding method of the conventional technology, the number of representative gradations to be assigned to a block is simply given according to the difference in gradation level within a block, so noise and edges cannot be distinguished, and in an image containing noise, There is a problem in that the code amount increases when trying to reproduce even edges with small tonal differences.

Further, in the encoding methods (1) and (2), the representative gradation is determined by comparing with the fixed thresholds T ₁ and T ₂ , so that an edge or a pattern with a small gradation change is reproduced with a low code amount. It was difficult.

On the other hand, in the encoding method of the above (3), since the variable threshold value is used, it is possible to reproduce an edge having a small gradation difference in a block having an average gradation of an intermediate level.
However, there is a problem that the code amount increases in an image including noise.

In addition, the conventional halftone dot photograph identification processing method is effective in increasing the compression rate when encoding a halftone dot image. However, in a normal natural image, even if a noisy area corresponds to a halftone dot block or a block with an edge and a non-halftone dot block, not only the halftone dot block is smoothed but also the resolution component deteriorates, which is good. No image was obtained.

In view of the above-mentioned conventional problems, an object of the present invention is to provide a block coding method for a multi-valued image capable of restoring an edge part having a small gradation difference while reducing the coding amount of noise. .

[Means for solving problems]

The purpose is to obtain a maximum gradation value and a minimum gradation value in the block of a multi-valued image divided into blocks composed of a plurality of pixels, and a maximum gradation value and a minimum gradation value in the block. Means for comparing the difference between the gradation value and a predetermined threshold value to obtain the number of representative gradations in the block, and the maximum gradation value and the minimum gradation value based on the representative gradation number. Means for obtaining a representative gradation value by quantizing the range, a difference value indicating a distribution range of the representative gradation value in the block, a reference value set in the distribution range, and Means for calculating a resolution component indicating which of the representative gradation values the gradation value of each pixel is, and the calculated reference value, the difference value, and the resolution component are respectively coded. Turned into
Further, a calculation unit that calculates the number of times the gradation value in the block crosses the average value of the gradation values of the pixels in the block, and a threshold value that sets the threshold value to a different value according to the calculation result of the calculation unit. Setting means is provided, the representative gradation value is obtained based on the threshold value, and the difference value and the resolution component are calculated.

[Action]

That is, in the present invention, by identifying the distribution state of the gradation value of the multi-valued image in the block, whether the multi-valued image in the block has random noise or the structural edge is It is determined whether or not there is, and the threshold value that determines the representative gradation value is changed according to the determination result, so that it is possible to reproduce a meaningful edge and suppress the coding of noise. be able to.

〔Example〕

FIG. 1 is a block diagram of an embodiment of a multi-valued image block coding system according to the present invention.

In the figure, multi-valued image data input from the terminal 1 is accumulated in the buffer memory 2 for one block line,
Then, the multi-valued image data X _ij is read from the buffer memory 2 block by block.

That is, as shown in FIG.
It is divided into blocks composed of × 4 pixels, and one of these
Multivalued image data X _{ij of the} block (shown in FIG. 2 (a))
Are sequentially read.

The gradation value change amount detection unit 3 receives the input multi-valued image data X
_The maximum gradation value L _max and the minimum gradation value L _min are detected from _ij and output.

Next, the intermediate value calculation unit 4 constituting the spatial distribution measurement unit 20 calculates and outputs the intermediate value (L _max + L _min ) / 2 based on the maximum gradation value and the minimum gradation value L _max , L _min. To do. For example, when the gradation value of the multi-valued image in the block is as shown in FIG. 2A, L _max is “63” and L _min is “51”, so the intermediate value is “57”. It will be. Then, the binarization unit 5 receives the multivalued image data X _ij from the gradation value change amount detection unit 3 and the intermediate value “5”.
7 ”and binarizes this multi-valued image data X _ij .
The result of this binarization is shown in FIG. The binarized data storage unit 6 stores the binarized data λ _ij for one block. Then, the number-of-changes counting unit 7 sequentially reads the binarized data λ _ij from the binarized data storage unit 6 in the horizontal direction and the vertical direction, respectively, and changes the number of changes R ₁ of “1” and “0” R _{1 respectively.} And R
Count ₂ . In the example shown in FIG. 3, the number of horizontal changes R ₁ is “4”, and the number of vertical changes R ₂ is “2”.

In this way, the spatial distribution within one block is measured by counting the number of changes R ₁ and R ₂ of “1” and “0” within the block.

Then, in the threshold selection unit 8, the counting results R ₁ and R ₂ from the change number counting unit 7 are input, the counting results R ₁ and R ₂ are input, and the counting results R ₁ and R ₂ are given in advance. If R ₁ ≧ S and R ₂ ≧ S by comparing with the threshold value S, the gradation number determination unit 9 is instructed to select a large value T _1H for the threshold value T ₁ for representative gradation number determination. .

On the other hand, if R ₁ <S or R ₂ <S, the gradation number determination unit 9 is instructed to select a small value T _1L for the threshold value T ₁ for the representative gradation number determination.

That is, when the number of changes in both directions is larger than the predetermined threshold value S, it is determined that the block has random noise, and the value of the threshold value T ₁ is increased to increase the probability of occurrence of one gradation quantization. . On the contrary, if the number of times is small in any one direction, it is determined that the structural edge exists in the block, the value of the threshold T ₁ is made small, and the probability of occurrence of 2-gradation quantization is made high. Is.

 This method is formulated as follows.

λ _ij = 0 X _ij ≧ (L _max + L _min ) / 2 λ _ij = 1 X _ij <(L _max + L _min ) / 2 thenT ₁ = T _1H elseT ₂ = T _1L (T _1H > T _1L ) Next, in the gradation number determination unit 9, the maximum gradation value L _max and the minimum gradation value in the block from the gradation change amount detection unit 3 are calculated. The difference value L _max −L _min is calculated based on L _min .

Then, the threshold value T _{1 for} representative gradation number determination selected based on the instruction from the threshold value selecting unit 8 is compared with the difference value to determine whether the representative gradation number is 1 or 2 or more. . Further, when it is determined that the number of representative gradations is 2 or more, the obtained difference value is compared with a _second threshold value T ₂ given in advance to determine whether the number of representative gradations is 2 or 4 or more. .

Subsequently, the representative gradation value determination unit 10 obtains the gradation in the block by linear quantization or non-linear quantization according to the obtained number of representative gradations.

That is, when the number of representative gradations is "1", as shown in FIG. 4 (a), the average value within the block is obtained and this value is set as the representative gradation value O _1, and the reference value L _A. Then, as shown in FIG. 5A, fixed values “0” are assigned to all pixels as the resolution components Φ ₁ and Φ ₂ .

When the number of representative gradations is "2", FIG.
As shown in, first, (maximum gradation L _max + minimum gradation L _min )
÷ 2 = L _M is calculated as an intermediate value and then the intermediate value
The average value of the gradation value of each pixel having the tone value in the range of L _M ~ maximum gradation value L _max, likewise, the intermediate value L _M ~ minimum gradation value L _min
And the average value of the range, and the respective average values are set to the respective representative gradation values P ₁ and P _2, and the difference value between the average values is calculated.
Find L _D. The reference value L _A is calculated as the average value of P ₁ and P ₂ (P ₁ + P ₂ ) / 2.

As shown in FIG. 5 (b), the resolution component is such that Φ ₂ is “0” in all pixels, and if the representative gradation is P ₁ , Φ ₁ = 0, the representative gradation is P ₂ Then, Φ ₁ = 1.

Further, when the number of representative gradations is “4”, FIG.
As shown in, first, the maximum gradation value L _max and the minimum gradation value L _min
Is equally divided into four, and the average value of the gradation values of the pixels having the gradation value within the range of 1/4 from the maximum gradation value L _max , and similarly, from the minimum gradation value L _min to 1 / An average value of gradation values of pixels having gradation values within the range of 4 is obtained. Then, the reference L _A is obtained as the average value of both average values. Further, the obtaining a difference value L _D between the two mean values, 3 equally divided between both average values.

The two average values thus obtained and the two equally divided values existing between the average values are set as representative gradation values.

The representative gradation values Q ₁ , Q ₂ , Q ₃ , Q ₄ are assigned as shown in FIG. 4 (c).

On the other hand, as shown in FIG. 5 (c), when the representative gradation value is Q ₁ , the resolution component is Φ ₁ = 0, Φ ₂ = 0, and the representative gradation value is Q ₂
, Φ = ₁ , Φ ₂ = 0, and when the representative gradation value is Q ₃ , Φ ₁ =
When 1, Φ ₂ = 0 and the representative gradation value is Q ₄ , Φ ₁ = 1 and Φ ₂ = 1.

By performing the above processing, the representative gradation value, the reference value, and the difference value in the block are calculated.

Then, the gradation value storage unit 12 stores the representative gradation value of the target block thus obtained.

The comparison unit 13 sets the multi-valued image data X _ij of the target block to 1
The pixels are read out again and compared with the representative gradation value stored in the gradation value storage unit 12, and the resolution components Φ ₁ and Φ ₂ corresponding to the representative gradation value with the smallest error are output.

The resolution components Φ ₁ and Φ ₂ are stored in the resolution component storage buffers 14a and 14b, respectively.

Then, the contents of this buffer 14a, 14b is the encoding unit 15a, 1
After being read by 5b and subjected to known MMR encoding processing to suppress redundancy, the resolution component storage unit 16a,
It is stored in 16b.

On the other hand, the reference value L _A and the difference value obtained by the representative gradation determination unit 10
L _D is variable-length coded by the code generators 11a and 11b and stored in the buffer memories 17a and 17b, respectively.

Then, when the processing for one screen is completed, L _A , L _D , Φ ₁ ,
The encoded signals of Φ ₂ are sequentially selected and read out via the multiplexer 18, and are transmitted to the receiving side in the above order.

In this embodiment, the difference value between the maximum gradation value L _max and the minimum gradation value L _min is compared with T ₁ , but the value to be compared does not have to be the difference value and the gradation value in the block Any value representing the amount of change in the value may be used, and for example, the difference value P ₁ −P ₂ of the representative gradations of the linearly quantized two gradations may be compared with the threshold value T ₂ .

Further, in the spatial distribution measuring unit, the average value of the maximum gradation value and the minimum gradation value is used as the binarization threshold value.
Other than this, the average value of the gradation values of all the pixels in the block may be used as the threshold for the binarization.

Furthermore, the threshold value T ₁ for determining the number of representative gradations that distinguishes between 1-gradation quantization and 2-gradation quantization is changed according to the spatial distribution measurement result. Similarly, T ₂ , T ₃ , ... can be changed, and the threshold value T ₁ can be changed according to the average value in the block, and the determination can be combined with other determinations.

〔The invention's effect〕

As described above, according to the present invention, a pixel in a block has random noise by measuring the spatial distribution of the gradation value for each block and determining the state of the gradation value of the pixel in the block. It is possible to improve the reproducibility of the essence and suppress the coding of noise because the threshold for the number of representative gradations is changed by determining whether there is a significant edge or not. It is possible to transmit a large image with a low code amount.

[Brief description of drawings]

FIG. 1 is a block diagram of an embodiment of a block coding method for a multi-valued image according to the present invention, FIG. 2 is an explanatory diagram of block division of a multi-valued image, FIG. 3 is an explanatory diagram of spatial distribution measurement, and FIG. FIG. 5 is an explanatory diagram of a method of calculating a representative gradation value, FIG. 5 is an explanatory diagram of resolution components, FIG. 6 is an explanatory diagram of linear quantization of a gradation range of an image signal, and FIG. 7 is a block gradation number. FIG. 8 is an explanatory view of coding information, FIG. 8 is an explanatory view of non-linear quantization of a gradation range of an image signal, and FIG.
FIG. 10 is an explanatory diagram of the variable threshold value, and FIG. 10 is a processing flow chart of the halftone dot block identification method. In the figure, 2 is a buffer memory, 3 is a gradation value change amount detecting unit, 8 is a threshold value selecting unit, 9 is a gradation number determining unit, 10 is a representative gradation determining unit, 11 is a code generating unit, and 12 is a gradation. A value storage unit, 13 is a comparison unit, 14 is a resolution component storage buffer, 15 is a resolution component encoding unit, 16 is a resolution component storage unit, 17 is a buffer memory,
Reference numeral 18 is a multiplexer, and 20 is a spatial distribution measuring unit.

 ─────────────────────────────────────────────────── ─── Continuation of the front page (72) Fumitaka Abe Inventor Fumitaka Abe 1015 Kamiodanaka, Nakahara-ku, Kawasaki City, Kanagawa Prefecture Fujitsu Limited (72) Inventor Hiroshi Ochi 1-2356 Takeshi, Yokosuka City, Kanagawa Prefecture Nippon Telegraph and Telephone Corporation Communication Laboratory (72) Inventor Makoto Kobayashi 1-2356 Takeshi, Yokosuka City, Kanagawa Japan Telegraph and Telephone Corporation Complex Communication Laboratory (72) Inventor Hisa Ibaraki 1-2356 Takeshi Yokosuka City, Kanagawa Prefecture Nippon Telegraph and Telephone Corporation (72) Inventor Minami Satoshi 386-1 Shimizu, Higashiyamato-shi, Tokyo 13 (72) Inventor Nakamura No. 2-4 Zenpukuji, Suginami-ku, Tokyo

Claims (3)

[Claims] 1. A means for obtaining a maximum gradation value and a minimum gradation value in a block of a multi-valued image divided into a block composed of a plurality of pixels, and a maximum gradation value and a minimum gradation value in the block. Means for comparing the difference between the value and a predetermined threshold value to obtain the number of representative gradations in the block, and the maximum gradation value and the minimum gradation value based on the representative gradation number. Means for quantizing the range of to obtain a representative gradation value, and a difference value indicating a distribution range of the representative gradation value in the block,
A reference value set within the distribution range, and means for calculating a resolution component indicating which of the representative gradation values the gradation value of each pixel in the block is, A block encoding device for a multi-valued image, which encodes each of the calculated reference value, difference value, and resolution component, wherein the gradation value in the block is the gradation of a pixel in the block. A calculating unit that calculates the number of times the average value is crossed, and a threshold setting unit that sets the threshold value to a different value according to the calculation result of the calculating unit are provided, and the representative gradation value is set based on the threshold value. A block coding apparatus for a multi-valued image, characterized in that the difference value and the resolution component are calculated. 2. The average value of the gradation values of the pixels in the block is an average value of the maximum gradation value and the minimum gradation value in the block. ) The block encoding device for a multivalued image described. 3. The multi-valued method according to claim 1, wherein an average value of gradation values of pixels in the block is an average value of gradation values of all pixels in the block. Value image block coding device.