JPS61217883A - Picture per-processing system - Google Patents

Abstract

PURPOSE:To obtain the binary picture data of high picture quality by dividing the Laplacian value with a certain reference value within a mixed picture and giving the selective contour emphasis only to the part of the character line drawing. CONSTITUTION:The Laplacian value Lxy5 stored in a register 110 is compared with the prescribed reference value T through a comparator 111. When the Lxy5 is larger than the Y, a demultiplexer 112 sends the picture data 3 on a notice picture element A to an adder 113. The adder 113 adds the value 6 obtained by multiplying the Lxy5 by the coefficient K through a multiplexer 114 to the data 3. Thus the contour emphasized data is supplied to a register 116. If the Lxy5 is smaller than the T, the demultiplexer 112 sends the data 3 as it is to the register 116. The data 7 processed by said operations is binary coded with the threshold value D of a dither matrix through a comparactor 117.

Description

[Detailed Description of the Invention] [1] Summary of the Invention The present invention calculates the Laplacian of each pixel when performing preprocessing on multivalued image data in which text and line drawings and gray scale drawings are mixed.
By comparing this with a predetermined reference value, the density change in the vicinity of each pixel is known, and when the density change is steep, it is determined that the pixel is a character line drawing, and outline enhancement is selectively applied to that pixel. .

[Industrial application field]

The present invention inputs multivalued image data in which a grayscale image and binary images such as characters and figures (hereinafter referred to as single character line drawings) are mixed, performs preprocessing on the multivalued image data, and then converts it into a binary image. Regarding the method of

[Conventional technology]

Conventionally, when expressing halftones using black and white binary values, a systematic dither method is usually used because of its gradation reproducibility and easy hardware configuration. In the systematic dithering method, halftones are expressed by the density of black pixels, so good images can be obtained for grayscale images. However, since the resolution deteriorates, if the original contains a mixture of grayscale images and text and line drawings, dithering the entire image will deteriorate the image quality of the text and line drawings.

In order to prevent image quality deterioration of text and line drawings, two types of methods have been proposed in the past. One is to identify the two images and separate them.
One method is to convert into values, and the other is to perform preprocessing and then
This is a method of converting it into a value.

The former method is as follows: (1) Institute of Electronics and Communication Engineers Technical Research Report IE81-57 (1
981) "Study of pseudo halftone reproduction considering text images"
There is. In this method, the difference between the maximum value and the minimum value of the multivalued image signal is determined for each small area of 4×4 pixels, and whether the image is a character line drawing or a grayscale image is determined based on whether this difference is larger than a predetermined reference value.

As for the latter method, (2) Journal of the Institute of Image Electronics Engineers, Vol. 10, No. 5 (198
1) There is "3. Halftone expression 3-1. Halftone formation".

As preprocessing, this method applies contour enhancement (Laplacian) to a multivalued image signal pixel by pixel, and binarizes the obtained image data using a systematic dither method.

Of these two methods, the +11 method improves the image quality of text and line drawings compared to dithered images, but has the disadvantage that the deterioration of image quality due to identification tags becomes noticeable in grayscale images.

In addition, method (2) has better image quality than method (1), but because the outline is emphasized over the entire image, the image quality of either the text line drawing or the gray scale image may deteriorate depending on the magnitude of the effect of the outline emphasis. The disadvantage was that it became more noticeable.

To explain method (2) above in more detail, (a) Using a multi-valued image signal, the density value of the pixel to be processed (hereinafter referred to as the pixel of interest) is set to A, and the pixel of interest and its surrounding eight pixels (hereinafter referred to as the reference pixel) are Let B be the average density of a total of 9 pixels.

(bl) The density value of the pixel of interest after contour enhancement is determined by the following equation. However, K is called the contour enhancement coefficient and is the weight of contour enhancement.

L dry A+K (A-B)...■
(C) For the entire image, (al~(b
Repeat the process l.

(d) After completing the above processing, binarize using a 4×4 pixel dither matrix.

Edge enhancement is performed as described above, but if the effect of edge enhancement is small, the text and line drawings will not be improved much.On the contrary, if the effect of edge enhancement is increased to improve the text and line drawings, it will not be possible to improve the text and line drawings. However, there is a drawback that a change in shading occurs, resulting in deterioration of the shading image.

[Problem that the invention seeks to solve]

The purpose of the present invention is to perform contour enhancement only on the text and line drawing portions, without applying contour enhancement to the portions of the gray image, as preprocessing for multivalued image data 2 (+!!), which contains a mixture of grayscale images and text and line drawings. Accordingly, it is an object of the present invention to provide a method for preventing deterioration of the image quality of a grayscale image when it is binarized using a systematic dithering method.

[Means for solving problems]

The present invention takes advantage of the property that the value obtained by performing a Laplacian operation on the multi-value image data of a character line drawing is larger than the value obtained by performing a Laplacian operation on the multi-value image data of a grayscale image, and compares it with a reference value. The outline is selectively enhanced only in the area.

[For production]

A Laplacian operation is performed on the multivalued image data of each pixel, and the result is compared with a reference value to determine the density change in the vicinity of the pixel of interest. Then, from this density change, it is determined whether the pixel of interest is a character line drawing or a grayscale image, and only if it is determined to be a character linework, the outline is emphasized, and if the density change is gradual, the pixel is determined to be a grayscale image, No contour enhancement is applied. Therefore, deterioration in image quality caused by applying edge enhancement to a grayscale image is prevented.

〔Example〕

Before a detailed explanation of the present invention, the systematic dither method

[Brief explanation of the drawing]

The same figure shows an example of a dither matrix of the systematic dither method, and the numbers within the multiplication of the number of columns 19 and the number j of rows are dark values. Now 1
6. When dithering a slightly quantized multivalued image signal, the dark value of the dither matrix is compared with the corresponding area (block) of the multivalued image. The density level Q of each pixel (X, y) is compared with each threshold value T8 of the dither matrix corresponding to it, and Q
When Xy≧T i j, the black level of the main pixel is “1”
and when Q,y<T,J, the white level is set to "0". Then, the entire image is binarized by repeatedly using the same dither matrix. This is a method of binarizing a grayscale image. Next, the present invention will be explained in detail with reference to FIG. This figure shows a histogram of values obtained by applying a Laplacian operation to the density of each pixel of a character line drawing and a gray scale drawing. Now, if we take the pixel of interest A and the reference pixel B as shown in FIG.
Laplacian LXy is determined by the following formula. L xy = (A-B+)+ (A-Bz) + (
A-Bt)+ (A-84)4(A-Bs)+(A-B
b)+(A-By)+(A-Be)...
・■Here, as shown in Fig. 3, find the Laplacian histogram for the grayscale image and the character line drawing, set the point where the two intersect, that is, the value T at which the frequencies are approximately equal, as the reference value, and divide the Laplacian LX into two. . That is, text and line drawings and grayscale images are distinguished from each other based on the magnitude of the Laplacian L XY with reference value T as the boundary. In this way, the present invention sets a single reference value T in advance,
Laplacian L per pixel...1! : Compare T, L
,,>T, the pixel of interest A is determined to be a character line drawing, the multivalued image data of the pixel of interest A is subjected to contour enhancement, and conversely,
, <T, the pixel of interest is determined to be a grayscale image, and no processing is performed on the multivalued image data of the pixel of interest A, so that the original data is maintained. This operation is repeated for the entire image, and the obtained image data is binarized using systematic dithering to reduce the deterioration in image quality of the grayscale image. FIG. 1 is a block diagram showing an embodiment of a preprocessing circuit for carrying out the present invention based on the above principle, and shows a circuit that performs preprocessing on a multivalued image signal and then binarizes the signal. In the figure, 200 is a circuit that performs Laplacian operation on multivalued image data of each pixel, 201 is a circuit that compares the obtained Laplacian with a reference value, and 202 is a circuit that selectively performs edge enhancement based on the comparison result. It is. The preprocessing circuit in the figure shows a circuit when the above-mentioned formula 0 (see P, 4) is converted as follows. Here, regarding the mutual positional relationship between the pixel of interest A and the reference pixel B, as shown in FIG. 4, eight surrounding pixels of the pixel of interest A are referred to as reference pixels B. L=A+K (A-B) [However, B indicates the image data of the reference pixel]=A+' ((A-Bl)+ (A Bt)+...+
(A+Ba) l...■ [However, ni'
-to/9) =A+to' + (A-A)+ (A-B,)+(A
B2)+...+(A-88) 1 In the above formula (2), the formula () is different from the above-mentioned formula 0, and represents the formula for calculating the Laplacian. First, a mixed image signal l of text and line drawings and grayscale images is input to the demultiplexer (DMPX) 100. The demultiplexer 100 sequentially writes data for each line into memories (MEM) 101, 102, and 103. After writing the image data into the memories 102 and 103, the multiplexer (MPX) 104 sequentially reads out 9 pixels in total, 3 pixels per line, as shown in FIG. Also multiplexer (MPX) 10
5, among the nine pixels read out by the multiplexer 104,
Only the pixels corresponding to the pixel of interest are selected and read out. Then, each time the processing of one line is completed, the memory 101゜1
Writing and reading of 02 and 103 are repeated. Now, assuming that the multivalued image data is read at a resolution of 8 pixels/mm and 8 bits/pixel, one line has 2048 pixels when considering a B4 size document. In this case, memories 101, 102, and 103 have a capacity that can store one line of data, and are 8 bit x 204
It consists of 8 words. The procedure for writing and reading from memories 101, 102, and 103 will be specifically explained below. To start binarization, first clear the memo January 01, and write the data of the first line to the memory 102 and the data of the second line to the memo January 03. At this time, the allocation of numbers for memories 101, 102, and 103 is in the state shown in FIG. Therefore, the multiplexer 104 reads the data of the memory 102 twice, for a total of 6 pixels, and reads out the image data for 9 pixels, including the data read from the memo January 03.On the other hand, the multiplexer 105 reads The image data in the memory 102 to which number 1 is assigned is read out. Then, the readout is repeated pixel by pixel, and when the processing for the first line is completed, the data in the third line is read out from the memory 102 to which number 3 is assigned. Write to O1. When writing of the data on the third line is completed, the number allocation of the memories 101, 102, and 103 becomes as shown in FIG. 5 (2). In this case, the memory 101.
02 and 103 respectively have the third. 1st. Since the data of the second line has been written, the multiplexer 104
Now, read out three pixels from each memory. Also, in multiplexer 1.05, memory 1 is allocated with number 1.
The image data of 03 is read out and the second line is processed. As described above, data is written to the memory assigned the number 3 every time the processing of one line is completed. When the writing is completed, the memory number allocation is changed from 3 to 2, 2 to 1, and 1 to 3, and the multiplexer 105
The image data is read from the memory to which number 1 is assigned. By repeating this operation for all lines, image data for the entire screen can be read out from the memory. The image data 2 of the reference pixel read out by the multiplexer 104 is subtracted by the subtracter (StlB) 107 from the image data 3 of the target pixel read out by the multiplexer 105 and input to the register (REG) 106. The result 4 is added pixel by pixel by an adder (ADD) 109. By repeating this 9 times for all reference pixels, the register (REG) 110 is set to Laplacian L, 5
is stored. The above is a circuit for calculating the Laplacian, and the present invention is achieved by adding the circuit described below to this circuit. The Laplacian value 5 in the register 110 is compared with a preset reference value (T) in a comparator (COMP) 111. As a result of the comparison, if the Laplacian LXy5 is larger than the reference value T, the demultiplexer (DMPX)
112 sends the image data 3 of the pixel of interest A to the adder (800) 113, and the multiplier (MOL) 114 multiplies the Laplacian L, . Add to image data 3. This allows the register (
Contour-enhanced data is input to REG) 1.16. On the other hand, when the Laplacian L, , 5 is smaller than the reference value T, the demultiplexer 112 outputs the image data 3 of the pixel of interest.
is sent to the register 116 as is. Therefore, the image data of the pixel of interest A is not subjected to any processing,
The original data is saved. The data 7 processed through the above operations is binarized by a comparator (COMP) 117 using a threshold value of a dither matrix. The above is an operation for one pixel.
By repeating the process for every pixel of a line, or by repeating every l line for every line, preprocessed image data covering the entire screen can be obtained. In the embodiment shown in FIG. 1, if a signal processing processor is used, the addition, subtraction, multiplication, and division calculation parts of the discrete circuits can be carried out by the processor's ALU (Arithmetic and U).
dLogic Unit) and software processing, the circuit can be made more compact. In the above embodiment, the 8 connected pixels shown in FIG. 4 were used for the Laplacian calculation, but this uses 4 connected pixels (in FIG. 3, the pixels indicated by B +, B z, B 5, and B q). You can also do that. [Effects of the Invention] According to the present invention, the Laplacian value in a mixed image is separated by a certain reference value, and only the text and line drawing portions are selectively enhanced, so that both the text and line drawings and the grayscale image can have good image quality. 2
value image data can be obtained. 4. Brief description of the drawings Figure 1 is a diagram illustrating the configuration of an embodiment of the present invention, Figure 2 is a diagram showing an example of a dither matrix, Figure 3 is a detailed diagram illustrating the present invention, Figure 4 is an attention-grabbing diagram. FIG. 5 is an explanatory diagram of how to select pixels and reference pixels, and a diagram of memory number allocation. In the figure, 200 is a circuit that performs Laplacian operation;
01 is a circuit that compares the Laplacian of each pixel with a predetermined reference value, 202 is a circuit that performs edge enhancement based on the comparison result, and T is a predetermined reference value. 86 to Al B1 86: The main pigment and the main polishing t/1 selection part 41! I Fist Rai> Nokari (1) (2) (3) (4)
(5) ,,. Memory 101 ■ -B [1] [I] Memo, G 102 B [=5=] B ■ [=1=]
Figure 5

Claims (1)

[Claims] means (200) for performing a Laplacian operation on the density of each pixel in a preprocessing circuit that performs contour enhancement on multivalued image data containing a mixture of character line drawings and grayscale images and then binarizes the data;
means (201) for comparing a predetermined reference value (T) with a value obtained by performing a Laplacian operation on the density of each pixel;
and means (202) for selectively applying edge enhancement to a pixel, and selectively applying edge enhancement to the pixel when it is determined to be a character line drawing based on the comparison result by the comparison unit. Image pre-processing method.