JP2575229B2 - Image processing device - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION (Industrial application field) The present invention is used in a facsimile apparatus or the like, and performs encoding of an image binarized by an error diffusion method or decoding of an encoded code sequence. The present invention relates to an image processing apparatus that performs encoding or decoding efficiently.

(Prior art) Conventionally, as a technique in this kind of field, the 35th IPSJ Transactions, “Compression of Dither Images by Arismetic Coding” (Showa 62), Kazuharu Toyokawa, P. 1983-1984 There was what was described in.

Conventionally, no encoding method has been disclosed that can encode an error diffusion binary image at a practical compression rate. This is because, in the error diffusion binary image, properties required for increasing the compression ratio in the conventional encoding method, such as run, periodicity, and Markov property between neighboring pixels, are not clearly shown between pixels.

As a method of encoding a binarized image by the dither method, the methods described in the above-mentioned documents have been announced. This method utilizes the fact that there is a strong correlation between pixels in the period of the size of the dither matrix.

(Problems to be Solved by the Invention) However, in the above image processing apparatuses, a probability model is constructed from a superficial correlation appearing between pixels. Therefore, in a pixel diffusion binary image in which properties such as run between pixels, periodicity, and Markov property between neighboring pixels are not clear, it is almost impossible to expect the effect of compression by encoding.

An object of the present invention is to provide an image processing apparatus which solves the problem of the prior art that the effect of compression by encoding cannot be expected.

(Means for Solving the Problems) In order to solve the above-mentioned problems, the present invention is based on binary image data generated from a gradation image by an error diffusion method.
Arithmetic code generating means for generating an arithmetic code based on the coding parameter and outputting a coded sequence corresponding to the binary image data (or an arithmetic code generating means for generating a coded sequence corresponding to the binary image data,
Image processing comprising an input of a coded sequence of a value image, an arithmetic code decoding unit for decoding an arithmetic code with the coding parameters of the coded sequence and generating binary image data corresponding to the binary image) In the apparatus, the following means are provided.

That is, in the present invention, a gradation value estimating means for estimating a gradation value of the gradation image from the binary image data and outputting an estimated gradation value, and the binary image data based on the estimated gradation value A binarization error estimating means for estimating a binarization error of a pixel of interest in the center and outputting an estimated binarization error value; and estimating a symbol appearance prediction probability from the estimated gradation value and the estimated binarization error value. And a symbol appearance probability estimating means for setting the symbol appearance prediction probability as the code parameter.

The tone value estimating means may be configured to estimate the tone value by smoothing by a weighted average of a ratio of white pixels and black pixels of neighboring pixels existing in the vicinity of the target pixel.

A first error estimating unit configured to estimate the binarizing error based on the estimated gradation value and a diffusion error from a neighboring pixel existing near the pixel of interest;
A second error estimating means for estimating a diffusion error to the neighboring pixels from the binarization error estimated by the first error estimating means, and correcting the diffusion error based on a predetermined diffusion error limit value First correcting means for performing correction, second correcting means for correcting the binarization error based on the average error of the pixel of interest, and the estimation based on the correction results of the first and second correcting means. Output means for outputting a binarized error value may be used.

The symbol appearance probability estimating means may be configured to calculate the symbol appearance probability from a probability density function of an error with respect to a true value of a value obtained by adding the estimated diffusion error value to the estimated gradation value.

(Operation) Since the image processing apparatus is configured as described above, the tone value estimating means, the binarization error estimating means, and the symbol appearance probability estimating means estimate the error diffusion process from the binary image data. And calculates the appearance probability of the binary symbol. The arithmetic code generation means configures an arithmetic code using the appearance probability of the binary symbol as an encoding parameter. This makes it possible to construct a model of the coding probability based on the structure model of the error diffusion binarization, and to make the probability estimation in the coding of the binary image generated by the error diffusion method appropriate. Therefore, the above problem can be solved.

(Embodiment) FIG. 1 shows an embodiment of the present invention and is a configuration block diagram of an image processing apparatus configured as an encoder.

The encoder 10 receives binary image data 10a generated from a gradation image by an error diffusion method, and estimates an error diffusion process by an error diffusion process estimating unit 11; And an arithmetic code generating means 12 for generating and outputting a coded sequence So corresponding to the binary image data 10a. Here, as the arithmetic code in the arithmetic code generation unit 12, an arithmetic code that can construct an arbitrary probability model without using a coding probability model depending only on a specific event (for example, a run length) is used.

The error diffusion process estimating means 11 has a tone value estimating means 11a for estimating a tone value of a tone image from the binary image data 10a and outputting an estimated tone value. On the output side of the gradation value estimating means 11a, a binarization error for estimating a binarization error of a target pixel in the binary image data 10a based on the estimated gradation value and outputting an estimated binarization error value is provided. Estimating means 11b is connected,
The output side of the binarization error estimation unit 11b is connected to the symbol appearance probability estimation unit 11c together with the gradation value estimation unit 11a. The symbol appearance probability estimation means 11c estimates a symbol appearance prediction probability from the estimated gradation value and the estimated binarization error value,
It has a function of setting the symbol appearance prediction probability as an encoding parameter. The output side of the symbol appearance probability estimating unit 11c is connected to the input side of the arithmetic code generating unit 12.

Further, the tone value estimating means 11a smoothes by a weighted average of the ratio of white pixels and black pixels of neighboring pixels (a, b, c, d shown in FIG. 2) near the target pixel, The image processing apparatus further includes a gradation value buffer configured to estimate the gradation value, and further stores the obtained estimated gradation value. The tone value buffer includes a first tone value buffer 11a-1 for one scan line including the target pixel, and a second tone value buffer 11 for one scan line of the scan line processed immediately before.
a-2.

A binarizing error estimating unit 11b for estimating the binarizing error based on the estimated gradation value and a diffused pixel from a neighboring pixel present in the vicinity of the target pixel;
-1 and 2 estimated by the first error estimating means 11b-1.
A second method of estimating a diffusion error from the binarization error to the neighboring pixels;
Error estimating means 11b-2 and first correcting means 11 for correcting the diffusion error based on a predetermined diffusion error limit value.
b-3, a second correction unit 11b-4 for correcting the binarization error based on the average error of the pixel of interest, and the estimation 2 based on the outputs of the first and second correction units. And output means 11b-5 for outputting a quantification error value.

Further, the symbol appearance probability estimating means 11c is configured to obtain the symbol appearance probability from a probability density function of an error with respect to a true value of a value obtained by adding the estimated diffusion error value to the estimated gradation value.

Operation (A) of image processing apparatus configured as above
(E) will be described.

(A) Operation of Tone Value Estimating Means 11a The neighboring pixels (a, b, c, d in FIG. 2) near the pixel of interest gs to be coded are accessed, and Find the tonal value. As shown in FIG. 3, the obtained estimated tone values are stored in first and second tone value buffers 11a.
-1, 11a-2. Each time the processing of one scan line is completed, the first gradation value buffer 11a-1
Are updated with the contents of the second gradation value buffer 11a-2.

 The gradation value estimation is performed as follows.

gs = 0 (when the target pixel gs is the neighboring pixel a, b, c, d, and the total number of black is 0) gs = gsmax / 4 (the target pixel gs is the neighboring pixel a, b, c, d Gs = gsmax / 2 (when the target pixel gs is 2 among the neighboring pixels a, b, c, and d) gs = (3/4) gsmax ( When the target pixel gs is the total number of black among the neighboring pixels a, b, c, and d, gs = gsmax (the target pixel gs is the total number of black among the neighboring pixels a, b, c, and d) (Case of 4) (1) gsmax; gradation maximum value The above noted pixel gs is smoothed in a nearby small area to obtain an estimated gradation value.

▲ ▼ = (▲ ▼ + ▲ ▼ + ▲ ▼ + ▲
▼ + gs) / 5... (2) where ▲ ▼: Estimated gradation value at neighboring pixel a ▲ ▼; Estimated gradation value at neighboring pixel b ▲ ▼; Estimated gradation value at neighboring pixel c ▲ ▼: Neighboring pixel Estimated gradation value at d This triangle is stored in the second gradation value buffer 11a-2 as the estimated gradation value at the target pixel gs. Assuming that the gradation value is almost constant in the minute region in the original image, the estimated gradation value is obtained by smoothing in the minute region near the target pixel gs. The nearby minute area is considered to be an area including a set of pixels that can affect the smoothing in the above equation (2). With respect to the change in the gradation value, the adaptation is performed according to the fifth term in parentheses in the above equation (2).

(B) Operation of the binarization error estimating means 11b From the neighboring pixels (a, b, c, d in FIG. 2) existing near a certain pixel of interest gs to be coded, the pixel of interest gs
Is calculated. As shown in FIG. 4, the obtained binarization error is represented by first and second error estimating means.
First and second error buffers 11b-1 in 11b-1 and 11b-2
a, 11b-2a. Each time one scan line is completed, the contents of the first error buffer 11b-1a are updated with the contents of the second error buffer 11b-2a.

The binarization error estimation processing is performed as follows, as shown in FIG.

When the process is started (step 20), first, the diffusion error error diffused from the neighboring pixels to the target pixel gs is: error = (7/16) · ea + (1/16) · eb + (5/16) · Ec + (3/16) · ed (3) where ea; estimated binarization error in neighboring pixel a eb; estimated binarization error in neighboring pixel b ec; estimated binarization error in neighboring pixel c ed The estimated binarization error at the neighboring pixel d (step 21).

Subsequently, the correction of the diffusion error value by the diffusion error limit value GSmax is performed as follows.

The range that the diffusion error value error can take is −GSmax / 2, ≦ error ≦ GSmax / 2. If the diffusion error value error is (error <−GSmax / 2) (step 22), the diffusion error error is set to −GSmax / Corrected to 2 (error = -GSmax / 2) (step 23). If the diffusion error value is (GSmax / 2 <error) (step 2
4) Correct the diffusion error error to GSmax / 2 (error = GSma
x / 2) (step 25).

Then, binarization is performed. That is, dgs = ▲ + error where dgs is the tone value after error diffusion used for binarization at the target pixel gs (step 26), and t is a binarization threshold (= gsmax / 2)
Then, if (t ≦ dgs), the target pixel gs is predicted as a black symbol, and if (dgs <t), the target pixel gs is predicted as a white symbol (steps 27 to 31).

Further, when the target pixel gs is predicted as a black symbol,
The estimated binary error value (dgs-gsmax) is stored in a second error buffer.
11b-2a. When the target pixel gs is predicted as a white symbol, the estimated binarization error value (dgs) is stored in the second error buffer 11b-2a.

Then, finally, the binarization error is predicted and corrected. A symbol of a binary image at the same coordinates where the predicted symbol s of the target pixel gs obtained by binarizing the tone value dgs is encoded.
Perform the following processing based on whether or not ss matches (step 3
2). That is, in the case of (s = ss), the process ends without performing the prediction correction. If s ≠ ss, a prediction error has occurred, and the estimation error is corrected as follows. When s = black, since the correct symbol is white, the estimation error is corrected by the average error (gsmax / 4) generated at the occurrence of the white symbol, and stored in the second error buffer 11b-2a. s =
In the case of white, since the correct symbol is black, the estimation error is corrected by the average error (-gsmax / 4) generated in the occurrence of the black symbol, and is stored in the second error buffer 11b-2 (step 33). ~ 35), end the process (Step 3)
6).

(C) Operation of Symbol Appearance Probability Estimating Means 11c dgs is a predicted value, and assuming a prediction error e, the true value is (d
gs + e). When dgs <t, the probability P (W) that dgs + e <t is satisfied (the prediction hits when the prediction symbol is white)
Is P (B) = 1−P (W) where p; e probability density function P (W); predicted appearance probability of white symbol P (B); predicted appearance probability of black symbol t ≦ dgs, t The probability P (B) that satisfies ≦ dgs + e (the prediction is hit when the prediction symbol is black) is Becomes p is statistically obtained in advance as a characteristic of the estimation error value e. The above equations (4) and (5) are functions of dgs, and d
The amount of calculation can be reduced by referring to the table from the value of gs.

(D) Operation of Arithmetic Code Generation Means 12 Arithmetic codes are formed using the appearance probabilities P (W) and P (B) of binary symbols obtained by the symbol appearance probability estimation means 11c as encoding parameters.

(E) Overall Operation of Each Function Unit As shown in FIG. 6, the order in which encoding is performed is the order in which a binary image is scanned in units of scan lines.

FIG. 7 is a flowchart showing the overall control of the encoding process, which will be described with reference to FIG.

When the process is started (step 40), first, the first tone value buffer 11a-1 and the first error buffer 11b-1a are processed (step 41). First gradation value buffer
11a-1 is p-gsbuf, and the first error buffer 11b-1a
Is p-ebuf, p-gsbuf (x) ← gsmax / 2 0 ≦ x ≦ xmax where xmax; x coordinate maximum value p-ebuf (x) ← 0 is set.

Y coordinate and y of scan line and x on scan line
The coordinate values are initialized to (x ← 0, y ← 0) (step 4
2). In addition, arithmetic code "C" and Augend "A"
Is initialized to (C ← 0, A ← 0) (step 43). After the initial setting, the tone value estimating means 11a, the binarization error estimating means 11b, the symbol appearance probability estimating means 11c, and the arithmetic code generating means 12 described above are sequentially executed (steps 44 to 47), and the x value is set to 1 It increases (step 48) and forms a loop until the x value reaches xmax, and when the x value reaches xmax (step 4).
9) The contents of the buffer are moved as follows (step 50).

p-gsbuf ← gsbuf p-ebuf ← ebuf where gsbuf; second tone value buffer 11a-2 ebuf; second error buffer 11b-2a Further, the x value is increased by 0 and the y value is increased by 1 (step 51). ),
Control is transferred to the process after the initial value is set. When the y value exceeds ymax, all processing ends (steps 52 and 5).
3).

 Hereinafter, experimental results showing the effects of this embodiment will be described.

The Pasco arithmetic code cited in the IEICE research report, "A Consideration on Arithmetic Coding of Multivalued Images", IT88-115, is used for binary symbols. In addition, the tone image area (person) of the facsimile test chart No. 1 and VOL4 of the Institute of Image Electronics Engineers of Japan is used as a tone image (image size: 512 × 51).
2, gradation value; 64).

This gradation image is subjected to an error diffusion method (diffusion coefficient; 7/16, 1/16, 5/16, 3/1 for each of the neighboring pixels a, b, c, d in FIG. 2).
In 6), binarization was performed, and the compression ratio in encoding was calculated. As a result, (the amount of information before compression; 262144 bits, the amount of information after compression; 81917 bits) was obtained, and the compression ratio was 3.2 times.

Next, an embodiment of decoding corresponding to encoding will be described.

FIG. 8 shows an embodiment of the present invention, and is a configuration block diagram of an image processing apparatus configured as a decoder.

In this decoder, the only means different from the encoder shown in FIG. 1 is an arithmetic code decoding means 12a for decoding an arithmetic code and generating binary image data 12b, and other means. Tone value estimation means 11a, binarization error estimation means 1
1b and the symbol appearance probability estimating means 11c have the same configuration as in FIG.

 Next, the operation will be described.

The same processing as the above-described coding is performed, except that the above-described coding is reversed by the arithmetic code decoding means 12a. The binary pixels already decoded by the arithmetic code decoding means 12a are input to the error diffusion process estimation means 11. In the error diffusion process estimating means 11, the symbol is decoded by the arithmetic code decoding means 12a using the appearance probability of the next symbol, and the decoded binary image data 12b is obtained. As a result, an excellent decoding rate can be obtained almost in the same manner as in the embodiment shown in FIG.

(Effects of the Invention) As described above in detail, according to the present invention, instead of constructing a coding probability model from a superficial correlation appearing between pixels, a gradation value estimating unit, a binarization error Estimation means,
And a symbol appearance probability estimating means, and constructing a coding probability model based on the structural model of error diffusion binarization, so that the probability estimation accuracy in coding is high,
The code can be configured effectively.

[Brief description of the drawings]

FIG. 1 is a block diagram showing the configuration of an encoder according to an embodiment of the present invention;
FIG. 2 is a diagram showing neighboring pixels, FIG. 3 is a diagram showing storage formats of first and second tone value buffers, FIG. 4 is a diagram showing storage formats of first and second error buffers, FIG. 5 is a flowchart of a binarization error estimation process, FIG. 6 is a diagram showing the order of encoding, FIG. 7 is a flowchart showing overall control of the encoding process, and FIG. 8 is another embodiment of the present invention. FIG. 3 is a configuration block diagram of a decoder of FIG. 10a, 12b ... binary image data, 11a ... gradation value estimating means,
11b ... binarization error estimating means, 11b-1, 11b-2 ... first
11b-3, 11b-4 ... first and second correction means, 11b-5 ... output means, 11c ... symbol appearance probability estimation means, 12 ... arithmetic code generation means , 12a
... Arithmetic code decoding means.

Claims (8)

(57) [Claims] An arithmetic code generator for generating an arithmetic code according to an encoding parameter based on binary image data generated from a gradation image by an error diffusion method and outputting a coded sequence corresponding to the binary image data. A tone value estimating means for estimating a tone value of the tone image from the binary image data and outputting an estimated tone value; Binarization error estimating means for estimating a binarization error of a pixel of interest in the value image data and outputting an estimated binarization error value; symbol appearance prediction from the estimated gradation value and the estimated binarization error value A symbol appearance probability estimating means for estimating the probability and setting the symbol appearance prediction probability as the encoding parameter. 2. The image processing apparatus according to claim 1, wherein the gradation value estimating unit smoothes the image by smoothing by a weighted average of a ratio of a white pixel and a black pixel of a neighboring pixel near the target pixel. An image processing device configured to estimate a tone value. 3. The image processing device according to claim 1, wherein said binarization error estimating means is configured to calculate the binary error based on the estimated gradation value and a diffusion error from a neighboring pixel existing near the pixel of interest. First error estimating means for estimating a binarization error; second error estimating means for estimating a diffusion error to the neighboring pixels from the binarization error estimated by the first error estimating means; A first correction unit that corrects the second binarization error based on a predetermined diffusion error limit value, a second correction unit that corrects the second binarization error based on an average error of the pixel of interest, Output means for outputting the estimated binarization error value based on the correction results of the first and second correction means. 4. The image processing apparatus according to claim 1, wherein the symbol appearance probability estimating means calculates the symbol appearance probability as an error of an error with respect to a true value of a value obtained by adding the estimated diffusion error value to the estimated gradation value. An image processing device configured to obtain from a probability density function. 5. An encoding sequence of a binary image generated by an error diffusion method from a gradation image is input, and an arithmetic code is decoded by an encoding parameter of the encoding sequence to obtain a binary image corresponding to the binary image. An image processing apparatus comprising arithmetic code decoding means for generating value image data, wherein a gradation value estimating means for estimating a gradation value of the gradation image from the binary image data and outputting an estimated gradation value; A binarization error estimating means for estimating a binarization error of a target pixel in the binary image data based on the estimated gradation value and outputting an estimated binarization error value; An image processing apparatus, comprising: symbol appearance probability estimating means for estimating a symbol appearance prediction probability from an estimated binarization error value and setting the symbol appearance prediction probability as the encoding parameter. 6. The image processing apparatus according to claim 5, wherein the gradation estimating unit smoothes the gradation by performing weighted averaging of a ratio of a white pixel and a black pixel of a neighboring pixel existing near the target pixel. An image processing device configured to estimate a value. 7. The image processing apparatus according to claim 5, wherein the binarization error estimating means is configured to determine the binarized error based on the estimated gradation value and a diffusion error from a neighboring pixel existing near the pixel of interest. First error estimating means for estimating a binarization error; second error estimating means for estimating a diffusion error to the neighboring pixels from the binarization error estimated by the first error estimating means; A first correction unit that corrects the binarization error based on a predetermined diffusion error limit value; a second correction unit that corrects the binarization error based on an average error of the pixel of interest; And an output unit that outputs the estimated binarization error value based on a correction result of the second correction unit. 8. The image processing apparatus according to claim 5, wherein said symbol appearance probability estimating means calculates the symbol appearance probability as an error of a true value of a value obtained by adding the estimated diffusion error value to the estimated gradation value. An image processing device configured to obtain from a probability density function.