KR0150164B1 - Quantization method and apparatus using error diffusion for image processing system - Google Patents

Abstract

In binary quantization, the binarization error is quantized and stored to use it to reduce the size of the storage means for storing the error, and to reduce the number of times of writing and writing to the storage means. Comparing binary images and binary images binarized by the conventional error-diffusion binarization with a radially averaged frequency spectrum, there is almost no difference at low frequencies, a slight difference at high frequencies, and low frequency, a human visual characteristic. It is sensitive to signal and insensitive to high frequency signal so that there is almost no difference in image quality.

Description

Quantization Method and Apparatus by Error Diffusion Method in Image Processing Apparatus

1 is a circuit diagram according to the present invention.

2 is a concrete circuit diagram of the error quantization apparatus 101 of FIG.

3 is a visual frequency characteristic diagram of a human being.

4 is a spread spectrum comparison diagram of a conventional 14-bit error spreading binary image (4A) of the evolution method and an 8-bit error binary image according to the present invention.

The present invention relates to a quantization method in an image processing apparatus, and more particularly, to quantization by an error diffusion method which reduces the number of storage means for storing errors by quantizing and storing the binarization error, and reduces the number of reading and writing to the storage means. It relates to a method and an apparatus.

In general, error diffision halftoning in an image processing apparatus is known to have an excellent ability to reproduce continuous gradation images as binary images. The error diffusion binarization method is a method of approximating an average brightness value with a continuous grayscale image by adding a binarization error around the continuous grayscale input pixel and binarizing it by a specific threshold value. Recently, various binarization techniques have been proposed, which are modified from error diffusion binarization. In particular, the boundary-weighted error-diffusion binarization limited by Eschabch emphasizes the boundary to provide a visually clear binary image. However, in the process of binarization by the boundary emphasis error diffusion binarization method, the binarization error increases in proportion to the boundary emphasis coefficient to increase the amount of information stored in the storage means, and the data bus of the storage means is limited, thus writing to the storage means and reading from the storage means The number of times increases. If the number of reads and writes is increased, a problem of lengthening the binarization process by the error diffusion binarization method becomes long.

Accordingly, an object of the present invention is to provide a method and apparatus for reducing the size of the storage means for storing the error and reducing the number of times of writing and writing to the storage means by quantizing and storing the binarization error.

Another object of the present invention is to compare the binary image binarized by the error diffusion binarization method with the quantized error and the binary image binarized by the conventional error diffusion binarization method in the radially averaged frequency spectrum, and the result is almost low frequency. The present invention provides a method and apparatus that exhibits no difference, shows a slight difference at high frequency, is sensitive to low frequency signals, which are human visual characteristics, and is insensitive to high frequency signals, and thus has almost no difference in image quality.

It is another object of the present invention to provide a method of reducing the storage means and reducing the number of input / output to the storage means without significantly affecting the image quality.

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

FIG. 1 is a diagram for explaining an embodiment of the present invention, and is a circuit diagram showing an example of quantizing and storing errors.

A storage unit 100 for storing the calculated error value after processing the input image and the binarization, and an error adder 103 for receiving and adding the error value output from the error storage unit 111 of the storage unit 100.

An adder (104) for adding an output of the error adder (103) and an input image value of the storage unit (100).

A modified image storage unit 106 for temporarily storing the corrected image output from the adder 104;

And a skeleton that multiplies by a predetermined coefficient to emphasize the boundary of the input image of the storage unit 100.

An adder (107) for adding the output of the skeleton (105) and the output of the modified image storage unit (106).

A binary quantizer 108 for binary quantizing the output of the adder 107;

An error generator (102) for generating an error from an output of the corrected image storage (106) and an output of the binary quantizer (108).

An error quantization device 101 quantizes an error output value of the error generator 102 and stores the error output value in the error storage 111 of the storage 100.

2 is a diagram illustrating the quantization of the error and the writing / reading of the quantized error according to the present invention, FIG. 3 is a visual frequency characteristic diagram of a human being, and FIG. 4 is a conventional 14-bit binary image and 8-bit binarization of the present invention. Spectrum comparison example of the image.

Therefore, a specific embodiment of the present invention will be described in detail with reference to FIGS. 1 to 4.

The process of binarization with error-diffusion binarization is based on the modified input pixel [m (x, y)] and the threshold value sum of the input pixel [i (x, y)] and the marginal error sum [e _s (x, y)]. The binarization value [b (x, y)] is determined by comparing [t (x, y)]. Here, the error [e (x, y)] is obtained by subtracting the corrected input pixel from the binarization value, and the error diffusion binarization is developed in the following equation (1). In FIG. 2, the upper 8 bits of the 14-bit errors calculated by the error generator 102 are selected by the selector 201 and stored in the error storage unit 211 through the buffer 202, and then temporarily stored. The unit 204 illustrates an example in which the lower 6 bits are added to the stored error during pixel calculation to make 14 bits.

In Equation (1), a bundle of a _ij is called an error filter (A), and the value limited by floyd is as follows.

As for the error filter, many modified forms have been published since floyd published in 1976. Proc, Soc, Inf Disp, 17. An Adaptive of PP75-77, but the calculation amount is not large, and Equation (2) is a hardware implementation. This is the most commonly used because it's easy.

m (x, y) = i (x, y) + e _S (x, y) .......... (3)

b (x, y) = step [m (x, y) = t (x, y)] ......... (4)

e (x, y) = m (x, y)-b (x, y) ... (5)

In the boundary emphasis error diffusion binarization method, a high pass filtering of the input pixel can be obtained by using a threshold value t (x, y) proportional to the input pixel as in Equation (6).

t (x, y) = Kt _0- (K-1) i (x, y) ........... (6)

t0 is a constant, 0.5 when the input pixel is distributed between 0 and 1, and K is the boundary emphasis coefficient. In the case of the boundary emphasis error diffusion binarization method, an error range is derived from Equation (1)-(6) as shown in Equation (7).

Kt _0- (K-1) i (x, y)-255 ≤ e (i (x, y)) Kt _0- (K-1) i (x, y) ..... (7)

In Equation (7), it can be seen that the binary error depends on the brightness value of the input pixel, and the error increases in proportion to the boundary emphasis coefficient.

In the present invention, when binarization is performed by the boundary emphasis error diffusion binarization method, when the binarization error [e (x, y)] is stored in Equation (5), the error is quantized and stored in the storage device. Since this method can reduce the size of the storage means, it is possible to optimize the size of the storage means, and to reduce the number of times of writing and reading of the storage means when the data bus of the storage means is limited. The present invention will be described in detail by Eschbach's boundary emphasis error diffusion binarization method as Threshold MOD- of 1993. J of Electronic Imaging, Vol 2 (3), PP185-192.

When the input image is 8 bits and the boundary enhancement coefficient is K = 4, the error distribution is -508 ≦ e512. In Equation (2), since the error diffusion filter coefficients are 1/16, 3/16, 5/16, and 7/16, the error sum has a real value. The correction image value is calculated by the above equation (3), and the error calculated by the above equation (5) is obtained by subtracting the binary value from the correction image. The error has a real value by the error diffusion filter coefficient value. Since the minimum value of the error diffusion filter coefficient is 1/16, the minimum fractional value of the error is 0.0001 when expressed in binary. The error is 1 bit in code and is less than 512 so that the integer part is 9 bits and the decimal part is 4 bits and 14 bits. Therefore, in order to store an error, a storage unit of 14 bits is required. And if the data bus of the storage means is 8 bits, one error must be written to the storage means twice to write an error to the storage means, and two reads must be performed to read the data. Therefore, one pixel is binarized by the error diffusion method. You have to read and write. According to the present invention, the error is quantized and stored in 8 bits, so that one error is stored in 8 bits, and when the data bus is 8 bits, binarization processing can be performed by the error diffusion method by two error reads and writes.

FIG. 1 is for explaining the boundary emphasis error diffusion method. The error is read from the storage unit 100, the error is passed to the error adder 103, the error sum 20 is calculated, and the input image is read. The sum 20 is added to the adder 104 and stored in the modified image storage unit 106. For the boundary emphasis, the input image is multiplied by the K-1 scalar 105 and added to the output of the corrected image storage unit 106 and the adder 107, and this value is compared with K * 128 for binary quantization in the binary quantization unit 108. Recording on the binary image section 109. The comparator 102 for generating an error is 14 bits by subtracting the binary value 30 from the corrected image storage unit 106. The error quantization device 101 quantizes the generator error of the comparator 102 to 8 bits and stores the result in the storage unit 100.

1 illustrates a specific method of quantizing the error quantization device 101 with reference to FIG.

In FIG. 2, the quantization apparatus 201 includes a first temporary storage unit 203, a buffer 202, and a second temporary storage unit 204 for temporarily storing an error, and store an error of the storage unit 100. As described above, the upper 8 bits of the total 14 bit error spread values output from the comparator 102 of FIG. 1 are stored in the error storage unit 211 through the buffer 202. When reading from the error storage unit 211, the error 8 bits read from the error storage unit 211 are stored in the upper 8 bits of the second temporary storage unit 204, and the lower 6 bits are zero. Then, the error diffusion value of 14 bits stored in the second temporary storage unit 204 is transferred to the error adder 103 for error summing.

FIG. 3 is a model of the frequency characteristics of human vision according to the method SULLIVAN used in Image Halftoning Using --- in J. of OPt, Soc, Am, A / Vol 10, NO, 8, August 1993. Like the diagram, human visual characteristics have low pass characteristics. That is, the low frequency error is easily seen, but the high frequency error is difficult for human to recognize.

4 is an example of comparing a binary image output by error diffusion binarization with an error of 14 bits and a binary image binarized by the present invention using a spread spectrum, and having a brightness value of 0 to 255 in an input image. ) 256 × 256 images. The results of the two methods agree in the low frequency region, do not show a slight difference in the high frequency region, and are negligible in the high frequency region due to human visual characteristics.

By quantizing and storing the binarization error as described above, the size of the storage means for storing the error can be reduced, the number of reading and writing to the storage means can be reduced, and binarization by the error diffusion binarization method using the quantized error. Comparing the binary image binarized with the conventional error-diffusion binarization with an annular mean frequency spectrum, there is little difference at low frequencies, slight difference at high frequencies, and human visual characteristics are sensitive to low frequency signals. Since it is insensitive to high frequency signals, there is almost no difference in image quality, which is effective in reducing the storage means and reducing the number of input / output to the storage means without significantly affecting the image quality.

Claims (2)

In a quantization apparatus of an image processing apparatus. A storage unit (100) for storing an error value calculated after the input image to the image processing and the binarization processing; An error adder 103 for adding and receiving an error value output from the error storage unit 111 of the storage unit 100; An adder (104) for adding an output of the error adder (103) and an input image value of the storage unit (100). A modified image storage unit 106 for temporarily storing the corrected image output from the adder 104; And a skeleton that multiplies by a predetermined coefficient to emphasize the boundary of the input image of the storage unit 100. An adder (107) for adding the output of the skeleton (105) and the output of the modified image storage unit (106). A binary quantizer 108 for binary quantizing the output of the adder 107; An error generator (102) for generating an error from an output of the corrected image storage (106) and an output of the binary quantization unit (108). And an error quantization device (101) stored in the error storage unit (111) of the storage unit (100) by quantizing the error output value of the error generation unit (102). In a quantization method of an image processing apparatus. A first step of storing the calculated error after the input image and the binarization and outputting the stored error; A second step of generating a modified image temporarily by adding the error sum of the first step and the stored input image value; A third step of emphasizing the boundary of the input image and adding the modified image to the binary quantization; A fourth step of generating an error from the corrected image and the binary quantization value; And a fifth process of quantizing the error output value and storing the error output value in the error storage unit.