KR20010048592A - Principal Distance Constraint Image Halftoning for Homogeneous Dot Distribution - Google Patents

Abstract

PURPOSE: An image quantization method is provided to obtain a good quality of binary image by allowing a distance between binary pixels to be maintained at a particular distance and uniformly distributing the binary pixels spatially. CONSTITUTION: A principal distance is determined to obtain a uniform dot distribution with respect to an input gradation, which is an ideal distance to be maintained between binary pixels. Minor pixel values are determined with respect to the minor binary pixels. In order to maintain a distance between he binary pixels in the principal distance, a minimum distance is calculated about the minor pixels out of the minor binary pixels. A binary threshold value is changed and determined based on the principal distance and the minimum distance. A binary pixel value is determined based on the changed threshold value. Then, each of y-positions of black and white dots in binary upper line is stored. And, each of x-positions of black and white dots in the same line is stored. The minimum distance is calculated by determining x-search range according to an input minor pixel value and an input pixel(x(m,n); replacing the x-search range with L; increasing the searched x-position on n-line.

Description

Image quantization method to obtain uniform distribution between dots of binary image by maintaining optimal distance {Principal Distance Constraint Image Halftoning for Homogeneous Dot Distribution}

The present invention relates to an image quantization method for obtaining a uniform distribution between dots of a binary image by maintaining an optimum distance. More particularly, the present invention relates to a binary image according to an input gray value in applying an error diffusion method. The present invention relates to a method of improving image quality of a binary image by uniformly distributing binary pixels by adjusting a binary threshold so that binary pixels of an image maintain a constant distance.

In general, an image with 256 levels of brightness between 0 and 255 is called a continuous grayscale image, and half toning is a method of expressing a continuous grayscale image with a binary output device using only 0 (black) and 255 (white). The image generated by the halftoning method is called a binary image. As an half-toning technique widely used in the related art, there is an error diffusion method.

The present invention is applicable to all binary and plural gradation output devices, and in particular, in the case of a binary output digital printer, the image quality of the output image can be improved. In addition, a binary output device such as a digital printer reproduces the gray level of the original image by a half-toning method, which is a technique of visually generating gray, which is an intermediate gray level, by arranging black and white dots on a continuous grayscale image. The image quality of the binary image depends greatly on the reproducibility of the gradation and whether or not the pattern of the binary pixels is visually unobtrusive. In particular, human vision is insensitive to high frequency components and very sensitive to low frequency components. Therefore, the distribution of binary pixels in dark areas or bright areas having a lot of low frequency components is very important for the image quality of binary images compared to intermediate gray scales having many high frequency components.

As such, since the binary image is represented by only two states of black spots and white spots, the distribution of points of the binary image also depends on a value existing among the black spots and the white spots. In other words, when the input image is larger than 128 (the absolute value corresponding to the middle value of the continuous gradation input image), since there are more white spots than black spots, the black spot is called a minor pixel. The distribution of phosphorus spots determines the quality of the binary image. On the other hand, when the input gray scale value is smaller than 128, the decimal pixel becomes a white point, and thus the image quality is affected by the distribution of the white point.

An error diffusion method as a representative method for generating a binary image is expressed as in Equations 1 to 3 below.

In Equation 1, x (m, n) represents the continuous gray scale input value in the pixel, and w (k, l) represents the error diffusion coefficient defined in the R region around the (m, n) pixel. e (m, n) represents the binarization error value after determining the binary pixel value in the (m, n) pixel. In the process of determining a binary image in the conventional error diffusion method, the binary threshold value T in Equation 3 mainly uses a fixed value. This conventional error diffusion method has a disadvantage of generating a visually poor pattern because the binary pixels do not have a uniform distribution.

Efforts have been made to improve the shortcomings of the conventional error diffusion method. As a representative method, the error diffusion method propagates the error value e (m, n) to surrounding pixels by a predetermined weight w (k, l). The binary threshold value T is propagated to the surrounding pixels by the weight already determined according to the input gray value and the binary pixel value to modulate the threshold value (Eschbach's Algorithm: USP 5,535,019) and the number of pixels There is a method (Marcu's Algorithm) to determine the position in advance and thereby determine the value of the binary pixel.

Eschbach's method (USP 5,535,019) basically does not use a binary threshold value corresponding to T in Equation 3 as a fixed value, and the error value in Equation 1 is the pixel value of the original image x ( It is a method of changing the T value using a method similar to changing m, n) to generate a modified input value u (m, n). In order to change the T value, an amount corresponding to an error value must be determined as in Equation 1, and a corresponding value is a value defined as an imprint value. The equation for the function for determining the transfer value is shown in Equation 4 to Equation 6 below.

Subscripts W and B in the above equations indicate that the value of the binary pixel in the (m, n) pixel is a white point or a black point. That is, the above equation is a process of determining the propagated transfer value Imprint Value according to the input pixel x (m, n) and the output pixel value at (m, n). The transfer value determined in this way is weighted and propagated to surrounding pixels, wherein the weights used are shown in the table below.

Transcription Propagation Weighting Factor in Eschbach Method location -6 -5 -4 -3 -2 -One 0 One 2 3 4 5 6 weight 0.045 0.23 0.5 0.68 0.82 0.91 One 0.91 0.82 0.68 0.5 0.23 0.045

The position in Table 1 shows the relative position in the x-axis direction (same line in the image) with respect to the position (0) where the current binarization is performed. The process of Equation 4 to Equation 6 for determining the transfer value with respect to the values of the input pixel and the binary pixel and the process of weighting and propagating the transfer value as shown in Table 1 are performed for all the pixels. The equation for modulating the binary threshold is shown in Equation 7 below.

In Equation 7, T (m, n) is a binary threshold used when determining a binary pixel value in (m, n), and T _imp (m-1, n-1) in Equation 7 is (m -1, n-1) means a transfer value stored in the pixel, and the last term means a cumulative value of the transfer value from the line to binarization to the previous pixel.

The method proposed by Marcu is to predetermine the position of the decimal pixel according to the gray value of the input pixel, and to binarize the current pixel when the binary pixel is searched by the predetermined order. The binary output value for may have an output value opposite to the decimal pixel value, and if the decimal pixel is not found until the search is completed, the output value for the current pixel becomes the decimal pixel value.

Example of fractional pixel search order ... 20 ... ... 19 12 21 ... ... 18 11 6 13 22 ... ... 17 10 5 2 7 14 23 ... ... 16 9 4 One 0 3 8 15 24 ...

Table 2 shows a search order for searching for the presence or absence of a few pixels among the pixels already binarized from the current pixel (0). At this time, the value indicating the end of the search is determined according to the gray value of the input pixel. That is, assuming that the search limit value for the input pixel value x (m, n) is 22, the search order searches whether there are decimal pixels at positions corresponding to 1 to 22, If found, the search is aborted and binarization is performed with values opposite to the fractional pixel. On the other hand, if no decimal pixel is found during the search up to 22, the output value for the current pixel becomes the decimal pixel value. Table 3 shows the limit values of the search determined according to the input gray level.

Limit value of search according to input gradation value Gradation One 2 3 4 5 6 ... 251 252 253 254 Search limit 148 111 92 79 70 63 ... 79 92 111 148

The Eschbach method changes the T value by using a method that defines a transfer value and the error value is similar to generating a modified input value u (m, n) by changing the pixel value x (m, n) of the original image. The Marcu method is a method of binarizing a binary value by searching a binarized pixel value according to a predetermined order when binarizing a current pixel by pre-determining a position where a small number of pixels exist according to a gray value of an input pixel.

SUMMARY OF THE INVENTION An object of the present invention is to make a binary image having good image quality by spatially uniformly distributing pixels by maintaining a distance between binary pixels.

Therefore, since the distance to be maintained between the binary pixels is determined by the pixel value of the continuous grayscale image, a process of defining and calculating a principal distance, which is an ideal distance to be maintained between the binary pixels, In addition, the object of the present invention is to perform binarization by changing a binary threshold value using an optimal distance and a minimum distance determined by calculating a minimum distance from an input pixel.

1 is an overall flowchart of an image quantization method for obtaining a uniform distribution between dots of a binary image by maintaining an optimal distance of the present invention;

2 is a flowchart of a decimal pixel determination unit of the present invention;

3 is a flowchart of a minimum distance calculation unit of the present invention;

4 is a flowchart of a binary pixel y-axis value storing unit of the present invention;

5 is a distribution diagram of the points of the Eschbach method, the Marku method, and the binary image of the present invention;

6 is a view showing an image by the Eschbach method,

7 is a view showing an image by the Marku method (Marcu) method,

8 is a view showing an image according to the present invention,

9 is a graph showing an optimal distance between binary pixels according to an input gray scale;

10 is a diagram illustrating an example of distribution of a binary image;

11 is a diagram showing a relationship between a minimum distance and an optimal distance.

The present invention for achieving the above object comprises the steps of determining a Principal Distance, which is an ideal distance that must be maintained between binary pixels to have a uniform distribution of points with respect to the input grayscale; Determining a fractional pixel value for the fractionalized pixel already binarized; Calculating a minimum distance with respect to a minor pixel among the binarized pixels so that the distance between the binary pixels maintains an optimal distance; Determining by changing a binary threshold value using an optimal distance and a minimum distance; Determining a binary pixel value from the changed threshold; Storing the y position of each of the black and white points of the upper line already binarized; By storing the x positions of the black and white points of the same line already binarized, we defined the optimal distance, which is the ideal distance to be maintained between binary pixels according to the input gradation, and to maintain the optimal distance. It is a characteristic that the uniform point distribution was realized by changing the threshold.

Hereinafter, the present invention will be described in detail with reference to the accompanying drawings.

FIG. 1 is an overall flowchart of the present invention. First, in step S100, a Principal Distance, which is an ideal distance that should be maintained between binary pixels in order to have a uniform distribution of points with respect to an input gray scale, is determined.

The ideal frequency that the binary image according to the gray scale should have is defined by Equation 8 below.

In this development, the inverse of the ideal frequency defined in Equation (8) is defined as a principal distance, which is an ideal distance to be maintained between binary pixels.

The optimal distance calculated in Equation 9 is used to change the binary threshold value T.

9 is a graph illustrating an ideal distance between binary pixels according to input gray levels.

Next, in step S101, a fractional pixel value is determined for a fractional pixel that has already been binarized. In the determination of the fractional pixel value, a pixel that has already been binarized so that the distance between the binary pixels maintains an optimal distance. The minimum distance is calculated for a minor pixel among them. At this time, the determination of the decimal pixel (black point or white point) is determined according to the input gray scale. The equation for the determination of the minority pixel is shown in Equation 10 below, and is determined by the method shown in FIG.

If the input pixel x (m, n) is smaller than 128, the decimal pixel becomes a white point, and if the input pixel x (m, n) is larger than 128, the minor pixel becomes a black point. The distribution of the minority pixels thus determined determines the image quality of the binary image.

The determining of the fractional pixel value may include determining whether the input pixel x (m, n) is smaller than the middle gray value (S110), and determining the fractional pixel as a black spot when it is not smaller than the middle gray value. (S111) and determining a small number of pixels as a white point when it is smaller than the intermediate gray scale value (S112).

In addition, storing the position of the binary pixel stores the position of the binarized pixels in order to calculate the distance of the decimal pixel among the binarized pixels from the pixel which is currently binarized. The present invention proposes a method of minimizing the amount of memory used by changing a two-dimensional position into one-dimensional information to store the positions of binarized pixels. In storing the two-dimensional position, as shown in Table 4, the information about the y-axis and the information about the x-axis are separately stored.

Example of saving y-axis position information for binary pixels in the upper line 3 5 -2 7 One 7 -3 … 7 -3 7 One

Table 4 shows the number of pixels that occur continuously in the upper line of the pixel currently binarized. For example, 3 means that 3 white spots occur continuously in the vertical direction from the line immediately above, and -2 means that 2 black spots occur continuously. This method has the effect of encoding and storing the position on the y-axis of the few pixels that are closest to the upper line as the one-dimensional memory without using the two-dimensional memory. Table 5 below shows the actual value y _off of the position on the y-axis of the nearest few pixels from Table 4. In storing the position of the binary image as a cumulative value, the type of the binary image may be displayed using a sign. 10 is an example of the result of the binary image is to code the position of y in the manner shown in Table 5.

Example of relative y-axis values from position-saved look-up table (LUT) value 3 5 -2 7 One 7 -3 … 7 -3 7 One Sunspot y value -4 -6 -One -8 -2 -8 -One … -8 -One -8 -2 100 point y value -One -One -3 -One -One -One -4 … -One -4 -One -One

The same position can be stored for the binary pixels on the same line as the pixel (m, n) in the same manner.

In the storing of the position of the pixel as described above (S105), the positions of the black and white dots which have already been binarized are stored. The storing of the position of the binary pixel is performed by storing the binary image b (m, n) in y. Storing at the m th value of the axis storage memory (S130), determining whether the stored binary image is a black spot (S131), and determining whether the y _off value is less than zero (S132). And reducing 1 to the m-th value of the y-axis storage memory when the y _off value is less than 0 (S133), and + m to the m-th value of the y-axis storage memory when the y _off value is not less than zero. Initializing to 1 (S134), determining whether the y _off value is greater than 0 if not black (S135), and if the y _off value is greater than 0 as the m-th value of the y-axis storage memory Step S136, and -1 as the m th value of the y-axis storage memory when the y _off value is not greater than zero. It performs the step including the initialization to (S137).

The y position storage is performed each time the (m, n) pixel is binarized. There are m y-axis storage spaces. At this time, after binarizing the pixel of (m, n), the binary state of (m-16, n) is stored at the position of m-16 of the y-axis storage memory. For this reason, the y-axis storage memory must store the y position of the decimal pixel on the upper line, which is stored from (m-16) to (m + 16) when calculating the threshold value at the pixel of (m, n). This is because the y position of the upper line is used.

Therefore, after the binarization of the (m, n) pixel, the (m + 1, n) pixel does not use the minority pixel y information of the upper line at the position (m-16) (the maximum search is in the x-axis direction). You can change the position information of ± 16) (m-16).

In Table 6, the gray part does not affect binarization of (m, n) pixels, so the y-axis cumulative value is updated in consideration of the binarization result of the current line, and the white part is 2 of (m, n) pixels. In calculating the true result, it does not update because it contains y information of the upper line. That is, each time the pixel is binarized, the binary result of the pixel before 16 on the x-axis is updated to the y-axis accumulated value. For reference, when the pixel to be processed becomes the horizontal end of the image, the binarization is performed to update all 16 pixels before the y-axis accumulated value. This allows the minimum memory to be used and requires memory to store the 16 binary states corresponding to the 16 white positions in Table 6.

In addition, the detailed calculation process of the step (S102) of calculating the minimum distance for the minor pixels among the binarized pixels in order to maintain the optimum distance between the binary pixels is shown in FIG. have. The process of calculating the distance, i.e., the minimum distance, from the position of the nearest stored few pixels is as follows. In order to calculate the minimum distance, the distances between the current pixel position and the decimal pixels are calculated and the minimum value of the calculated distances is selected. At this time, the object of distance calculation calculates the distance only for a limited pixel according to the x-axis coordinate values. Table 7 shows the range of storage memory for which distances should be calculated for input pixels.

In Table 7, int (r) represents an integer value of the optimum distance already calculated for the input gray scale. That is, in order to calculate the distance of the minority pixels existing in the upper line, after calculating the position value on the y-axis of the minority pixels as described in Table 5 from the position of the memory in the range of int (r) from the current pixel, The minimum can be found by calculating the Euclidean distance. In the method of calculating the minimum distance, Euclidean distance is precomputed to find the position of an item by direct calculation and index it rather than comparing or scanning or indexing each item in the table. Can be used to save the calculation process and omit the calculation process. Table 8 below shows the LUT stored Euclidean distance.

Used distance storage LUT Memory x-axis position Extraction position of the fractional pixel Street 0 One One 2 2 3 3 … … 50 50 One One 1.14 2 2.263 3 3.162 … … 50 50 … 16 One 16.03 2 16.12 3 16.27 … … 50 52.49

Since the minimum integer including the maximum value of the distance calculated in Equation 9 is 16, the x-axis position of the memory in Table 8 has a value up to 16. The range of position values on the y-axis is also up to 50. The reason why the LUT is not defined for a value above 50 is that if the maximum value of the x axis value is 16, the y value can be approximated to about 50 if the value is 50 or more.

In addition, the step (S106) of storing the x position of each of the white point and the black point of the binary pixel has the same size as the horizontal size of the original image because the y position should have information of the binary pixel in the upper line binarized. While storage space is required, the x position storage unit can store the positions of pixels (0, n)-(m-1, n) for (m, n) pixels, and thus can store them in one variable. That is, the value changes every time the (m, n) pixel is binarized.

As described above, calculating the minimum distance (S102) includes determining a search x range according to the input pixel x (m, n) (S120), replacing the x range with -L (S121), and a line. Increasing the search x position on the image (S127), determining whether the x range is smaller than the added L (S122), and using the value stored in the binary pixel y coordinate if the x range is smaller than L. Obtaining a relative position value on the y-axis from a position-stored look-up table (LUT) (S123), and calculating the Euclidean distance d using the distance LUT using the search position x and y-axis coordinate values. Step S124, determining whether the minimum distance d _min is greater than the distance d, replacing the minimum distance d _min with the distance d, if larger, S126, and searching position x is greater than L and when compared to the d _min and x _min obtain in step (S128) of determining the distance of the nearest decimal pixel finally It is.

In operation S104, the binary threshold value is changed by using the optimal distance and the minimum distance, and in step S105, the binary pixel value is determined from the changed threshold value.

Referring to the threshold value change process, in order to maintain a constant distance between binary pixels in a binary image, the closest of the few pixels calculated from the optimal distance and the minimum distance calculation process, which are already defined ideal distances for the input grayscale, are described. Binarization is performed after changing the binary threshold T in Equation 3 using the distance. The change in the binary threshold at this time is illustrated by Figure 12 below.

In Equation 11, when the input pixel value is greater than or equal to 128, when the minimum distance d _min is greater than the optimal distance r, the binary output value at (m, n) must be binarized to a black spot to satisfy the optimal distance. If d _min is less than the optimum distance, the binary output for (m, n) pixels should be at the white point. The equation of change of the binary threshold is shown in Equation 11 below.

In Equation 11, A is any constant value between 10 and 25.

After changing the binary threshold value T by using the optimum distance and the minimum distance, binarization is performed by the calculation of Equation 12 to Equation 14 below.

The binarized pixels store the y-axis value of the binary pixel according to the process of FIG. 4.

According to the present invention described above, since the pixels are evenly distributed over the entire area of the binary image, and do not have a constant pattern as in the Marcu method, it has a good visual effect and can be applied to all binary and multi-gradation output devices. In particular, in the case of a binary output digital printer, the image quality of the output image can be improved. In FIG. 5, the 256x256 size image having the gray level 250 is binarized by the Eschbach method, the Marcu method, and the method according to the present invention. In addition, the method of the present invention shows that the binary pixels are uniformly distributed evenly over the entire area. FIGS. 6 to 8 illustrate an Eschbach method for a general image mainly having bright gray values. The result of binarization by the Marcu method and the method according to the present invention, as shown in the distribution of points in the area near the forehead. It can be seen that the distribution of pixels is uniformly distributed without a specific pattern, and the result of the error diffusion method for the uniform distribution of the binary image has an improved image quality compared to the conventional binary image.

Claims (8)

Determining a principal distance, which is an ideal distance that should be maintained between binary pixels to have a uniform distribution of points with respect to the input grayscale; Determining a fractional pixel value for the fractionalized pixel already binarized; Calculating a minimum distance with respect to a minor pixel among the binarized pixels so that the distance between the binary pixels maintains an optimal distance; Determining by changing a binary threshold value using an optimal distance and a minimum distance; Determining a binary pixel value from the changed threshold; Storing the y position of each of the black and white points of the upper line already binarized; And storing the x position of each of the black and white points of the same line already binarized. The method of claim 1, wherein the optimum distance in the step of determining the optimum distance Error diffusion method for the uniform distribution of binary images according to the distance constraint characterized in that is applied. The method of claim 1, wherein calculating the minimum distance Determining a search x range according to the input decimal pixel value and the input pixel x (m, n); replacing x with -L, increasing the search x position on line n, determining whether the x range is less than the added L, If the x range is less than L, using the values stored in the binary pixel y-coordinates to obtain the y-axis coordinates of the decimal pixel as a relative y-axis value from the look-up table lookup table (LUT), Calculating Euclidean distance d using distance LUT using search location x and y-axis coordinates, Determining whether the minimum distance d _min is greater than the distance d, If large, substituting the distance d _min for the distance d, If the search position x is greater than L, comparing the _{min min} and x _min and finally determining the distance of the nearest few pixels, error diffusion for uniform distribution of the binary image according to the distance constraint Way. The method of claim 1, wherein the storing of the position of the binary pixel is performed. Storing the binary image b (m, n) as the m th value of the y-axis storage memory; Determining whether the stored binary image is a sunspot; Determining whether y _off is less than 0 in the case of sunspot; if the value of y _off is less than 0, decreasing 1 to the m-th value of the y-axis storage memory; initializing it to +1 as the mth value of the y-axis storage memory when y _off is not less than 0; Determining whether the y _off value is greater than 0 when it is not a black point, incrementing 1 to the m th value of the y-axis storage memory when the y _off value is greater than 0; and a y-initializing value to -1 as the mth value of the y-axis storage memory when the y _off value is not greater than 0. The method of claim 1, wherein the threshold value change equation is determined in the step of determining the binary pixel value. (A is any constant value between 10 and 25), After changing the binary threshold T using the optimal and minimum distances, Error diffusion method for the uniform distribution of binary images according to the distance constraint characterized in that is applied. The method of claim 1, wherein the storing of the position of the binary pixel comprises storing a two-dimensional position of a binary image, changing a stored binary image into one-dimensional information, and storing the accumulated binary value as a cumulative value. Error diffusion method for uniform distribution of binary images according to distance constraint, characterized in that to minimize. The uniform distribution of the binary image according to the distance constraint of claim 1, wherein the calculating of the distance of the binary image comprises calculating a search range of a memory according to an input gray value and calculating a distance only for a limited pixel. Error diffusion method for 8. The binary method according to claim 7, wherein the method for calculating the distance for only the limited pixels is based on the search position of the memory as the x-axis and the distance is calculated in advance for the y-axis value and stored in the LUT. Error diffusion method for uniform distribution of images.