JPH01134577A - Image processor - Google Patents

Abstract

PURPOSE:To magnify and contract even an image obtained through pseudo halftone processing with reduced deterioration in image quality by producing multi-level data from binary data on an original image and applying a process where said multi-level data is converted into the density of a black picture element for the entire surface of the original image. CONSTITUTION:The multi-level data corresponding to the number of reproduced images are decided from the binary original image data via an original image data temporary memory part 100 and a multi-level data deciding part 200. The multi-level data is binarized by a binarization processing part 300. At the same time, a control part 400 controls the difference in picture element number between an original image and its reproduced image. For instance, the linear density conversion is carried out with a conversion rate r/R. In this case, the calculation of the multi-level data is carried out once per R times of the reference clock received from the part 400. While the input of the original image data is carried out once per (r) times of the clock. Then the images are magnified and contracted by binarizing the obtained multi-level data.

Description

[Detailed Description of the Invention] [Field of Industrial Application] The present invention performs image processing such as scaling and rotation on an original image expressed in binary data such as white and black, and expresses it in binary data. The present invention relates to an image processing apparatus for obtaining reproduced images, and particularly relates to an image processing apparatus suitable for obtaining reproduced images of high quality with a simple configuration.

[Conventional technology]

Using linear density conversion as an example, we will describe an example of a processing method used in a recent image processing device. Now, when enlarging the original image to 3/2, pixels of the reproduced image are allocated to 2/3 pixel intervals of the original image. The relationship between this original image and the reproduced image is shown in FIG. Here, each pixel of the original image is represented by P (xty) - each pixel of the reproduced image is represented by Q (xt y). Furthermore, the starting points P (0, O) and Q (0, 0) of one image are the same position.

Here, Q (2, O), Q (0, 2), etc. are the values of the original image P (3, O) and P (0, 3), respectively, but Q (1, 1), etc. Pixels located between the pixels of the image must be determined from the neighboring P(x, y).

This determination method is generally called interpolation processing, and includes the eyelid sum method,
The nearest neighbor method is widely known. Known examples include, for example, the document [Hiroshi Masaaki: Performance evaluation of various enlargement/reduction methods for binary images and processing speed improvement method, Transactions of the Information Processing Society of Japan, Vol. 26. No. 5 (September 1985)
pages 920 to 925].

Now, as shown in FIG. 3, P(t) on the original image.

j), P (i+1e j), P (t, j+1).

Q(m) surrounded by four points P(i+1+j+1).

Each method of interpolation processing will be explained for the case of determining n). The logical sum method is P (is j) y P (t+1゜j)+ p (it j+t)t p (++to j
In this method, if even one pixel in +t) is black, Q(men) is black. The nearest neighbor method is a method in which the value of the closest pixel among four pixels is set to Q (m, n), and if there are multiple closest pixels, the value is determined according to a specific priority order.

When linear density conversion is limited to image reduction, 1/2.2/
At a specific magnification such as 3, a method called thinning processing is also widely used. This processing is a method of thinning out data at specific coordinates from the original digital binary image data (hereinafter referred to as the original image) according to the reduction ratio. Specifically, for example, if you reduce the original image to 172, scan line 1
When reducing every other book to 2/3, the data is sampled by thinning out every three scanning lines one by one. - As an example, FIG. 4(b) shows the result of reducing the image shown in FIG. 4(a) to 1/2 using this method.

In this figure, the pixels shown with diagonal lines represent black pixels, and in the images in the figure, the direction of the arrow is the main scanning line direction,
The direction perpendicular to the arrow is the sub-scanning line direction. Figure 4 (,
) P (X? y) (main scanning line direction X.

If the data of the reproduced image obtained as a result of thinning in the sub-scanning direction y) is expressed as Q(x, y), then this method is
), (2, l>, (4, 1)...

(o, 2), (2, 2), (4, 2)...
・Q (0,0)=P (0,O) Q (0,1)=P (0,2) Q (0,2)=P (0,4) Q (1, This is a method in which the reproduced image data Q (x+y) is determined as 1)=P (2, 2).

On the other hand, in (Japanese Patent Application Laid-open No. 61-35073), when linear density conversion is performed on a binary image generated by a systematic dither method, the dither matrix used in the binarization is used as a unit, and the black color of each dither matrix is A method has also been proposed in which the density of each dither matrix is determined based on the number of pixels, and this value is binarized using a dither matrix with a different number of pixels. This method will be explained below. For example, consider a case where the binary image shown in FIG. 5(a) is an image binarized using a dither matrix of 4 pixels*4 pixels, and is enlarged 574 times. Divide the image in Figure 5(a) into blocks of 4 pixels x 4 pixels, count the number of black pixels in each block, and consider that value as the density of each pixel in the block. ) image data is obtained. In this case, the number of pixels is multiplied by 5/4 in both the vertical and horizontal directions for each primary block, where the density can take values in 17 steps from O to 16, to obtain the density data shown in FIG. 5(e). Then, it is binarized using a 5 pixel x 5 pixel dither matrix shown in FIG. 5(d), and
) to obtain a binary image.

Similarly, image rotation processing has conventionally used the OR method or the nearest neighbor method in order to determine the pixels that will be located between the pixels of the original image as a result of the processing.

[Problem that the invention seeks to solve]

Most of the conventional methods described above are aimed at processing ordinary binary images such as single characters and figures, and do not take into account processing of pseudo-halftone images such as photographs. Therefore, in the case of a pseudo-halftone image in which white and black pixels frequently switch, the logical sum method ends up painting most of the image black. Also, in the & neighborhood method, missing parts are noticeable when downscaling,
When enlarged, the smoothness of the image deteriorates. On the other hand, thinning processing reduces the number of pixels by extracting only part of the data. Since no compensation is made for the data missing due to thinning, the quality of one image is severely degraded, and for example, supplementary lines and minute points are missing. The pseudo halftone image is *
This method is not suitable because the image is formed using small dots. Furthermore, when a binary image formed by systematic dithering is reduced by this method, the period of the dither matrix and the thinning period interfere with each other, thereby degrading the image quality by -0. On the other hand, (Unexamined Japanese Patent Publication No. 6
1-35073) is intended only for images obtained by systematic dithering, and does not consider other images. Furthermore, since this method determines the density for each dither matrix such as 4 pixels*4 pixels, the resolution is significantly reduced. Furthermore, since this method is limited in the conversion rate that can be processed, the magnification is limited in enlargement/reduction processing, and the angle is limited in conversion processing.

The object of the present invention is to apply not only normal binary images but also images generated by pseudo halftone processing.

It is possible to perform conversion processing such as enlargement and reduction twice at an arbitrary conversion rate and obtain a reproduced image with little deterioration in image quality.
The object of the present invention is to obtain an image processing device.

[Means for solving problems]

The above purpose is to generate multi-value data from the binary data of the original image, and perform binarization processing on the entire yK image using a method of converting the multi-value data into black pixel density. This can be achieved by obtaining reproduced image data.

To generate multivalued data, refer to multiple pixels on the original image corresponding to the reproduced image, for example, when determining Q (m, n) in FIG. 3, refer to the pixels near (Is j) of the original image. This can be achieved by doing this.

[Effect]

The operating principle of the present invention will be explained below. Now, when performing linear density conversion on a binary original image of X1 pixels - Y4 pixels at a conversion rate r/R to obtain a reproduced image of X pixels * Y pixels, the present invention performs the following two steps. Perform processing.

(1) From the binary data for X1 pixels*Y1 pixels, generate multi-value data for X pixels*Y pixels corresponding to each pixel of the reproduced image.

(2) Binarize the multi-value data for X pixels*Y pixels to obtain binary reproduced image data for X pixels*Y pixels.

First, the process (1) will be explained. The process (1) involves simultaneously performing a process of determining multi-value data for one pixel by referring to a plurality of binary data and an interpolation process of image data.

In order to determine the multivalued data of pixel number X2+72, a plurality of pieces of original image data are referred to. Now, assuming that the starting point (0, O) of the original image and the reproduced image is a common position, the pixel number Xt+71, which is the center of the referenced original image, is defined by the following equation.

X 1 = [R/ r * x 2 ] Y 1 = [
R/r*72] Here, [x] represents the largest integer not exceeding X. For example, when reducing the original image to 273, X 1 = [3/2 * X 2 ]yt = [3/
2*y2]. The positional relationship between X1yY1 and X21V2 at this time is shown in FIG.

In the present invention, the multivalued data S(x,y) is determined by the number of black pixels in the original image data of m*n pixels surrounding it. Here, m and n are arbitrary. For example, if the number of pixels to be referenced is 4 pixels * 4 pixels, s in Figure 6
The most basic method for calculating (xzt yz) is 1.1 J wt However, x3=x, -2 'l3=1.2 This formula is 16 This means that the number of black pixels in the pixel is multivalued data 5 (xztyz). Therefore, the possible value of multivalued data S is 0
There are 17 stages from 16 to 16.

In actual processing, the 166 pixel values to be referenced are weighted based on the distance between the multivalued data S and each original image data P.

In this case, the weighting coefficients α and j are 4*4 matrices, and are functions of the relative positions of the multivalued data and the original image data. The relative position between the position (xszyz) of this multivalued data and the referenced original image data is determined by ΔX and Δy in the figure. Therefore, multi-value data 5 (X2*ya
) is determined as 1-I J-but, x3=xl-2 Ya=)rt 2 .

By repeatedly executing this process, it is possible to obtain multi-value data for the same number of X pixels*Y pixels as the reproduced image.

Next, the principle of the process of determining binary reproduced image data Q from multi-value data S will be explained using two main types of methods as examples. Note that in the present invention, various conventionally used binarization processing methods can also be applied to this part of the processing.

The first method is a method in which binary data is determined so that the cumulative total of rounding errors that occur when multilevel data is binarized is always minimized. Figure 7(a) shows the concentration 0 to 16.
This is an example of multivalued data S(X*y) with 7 gradations. When each pixel is binarized, 1≦5(x
When ey)=15, a rounding error ε occurs. Therefore, by extracting this error ε and adding it to the multivalued data of pixels that have not yet been subjected to binarization processing, the cumulative total of errors can be minimized. For example, each pixel in FIG. 7(a) is set to a threshold T=
When binarized at 8, the pixel at coordinates (1, 1) becomes S (
Since 1,1)=7 and 7≦8(=T), binary data Q (
1, 1)=0, and a rounding error t=7 occurs. Therefore, this ε is added to the multivalued data S (2,1) when the next pixel (2,1) is binarized. Then, 1 is originally S(2°1)=5, but it becomes 12 by adding the error 7. Therefore, it is determined that Q (2,1)=1. In other words, if a pixel with density 7 is white, the rounding error is added as the density of the next pixel, making it easier for the primary pixel to be determined to be black. Therefore, by repeating this process, the density can be converted into the density of black pixels.

The result of performing this process on the entire figure 7(a) is shown in Figure 7.
Shown in Figure (b). Note that it is also possible to disperse errors caused by binarization into multivalued data of a plurality of neighboring pixels.

In this case, for example, the pixel number (1°l) in FIG. 7(a)
The error 7 generated in
2) It is distributed to (2, 2), etc.

A second example of the binarization method is a method in which the number of black pixels to be allocated to a certain local region is determined based on the sum of the densities within a certain local region, and the pixels are arranged in descending order of density. For example, a case will be described in which multivalued data of 17 gradations from 0 to 16 shown in FIG. 8(a) is scanned using a scanning window corresponding to one pixel per pixel. The scanning window is A in the figure.
, the sum of the densities of the four pixels in the window is Sm
, the maximum value that the multivalued data of each pixel can take is C, and the number N of black pixels to be allocated to these four pixels and the remainder n are determined according to the following formula.

STr,=N*C+n In this example, Sm=45. From C=16, N=2゜n=
It becomes 13. Therefore, one pixel out of the four pixels in the scanning window is
6.1 pixel is replaced with 13, and the remaining 1 pixel is replaced with 0. The results are shown in FIG. 8(b). As a result of the processing, most pixels can be separated into 0 and 16 without changing the total density in the window. When this process is executed while scanning the window one pixel at a time, each pixel becomes
The replacement process will be performed a total of four times. The result of performing the replacement process once on the original multivalued data S (xt y) is
1 (xt y), S a
(xt y) is obtained. So next this 84 (xt
y) is binarized. In many cases, 5a(xpy) is O or 16, so no rounding error occurs due to binarization.However, 54(XIY) is 1≦5o(x
py)≦15, the resulting error E is added to the density sum Sm when the scanning window reaches the next position E. This is similar to the method 1 described above, in which the rounding error caused by binarization can always be minimized while converting the density to the density of black pixels by one or more processes for minimizing the cumulative total of errors.

Through the processes (1) and (2) described above, the binary original image data P(x, y) is subjected to linear density conversion to obtain the binary reproduced image Q(xpy).

〔Example〕

The present invention will be described in detail below with reference to one drawing.

FIG. 1 shows an example of the basic configuration of the present invention. First, 101 is a signal line for inputting binary original image data, and 100 is a temporary storage section for storing original image data for a plurality of scanning lines.

200 is a multi-value data determination unit that calculates multi-value data for one pixel from the original image data in the original image data storage unit 100, and 300 is a unit that binarizes the multi-value data output from the multi-value data determination unit. This is a multi-value data determining unit. The binarized data is output from the signal line 391 as reproduced image data. In other words, the original image data storage unit 100 and the multi-value data determination unit 200 determine a number of multi-value data from the binary original image data according to the reproduced image.
The process (1) described in the [Operation] section is executed, and the multi-value data determination unit 300 performs the binarization process (
[Operation] The process (2)) described in item '' is executed. Furthermore, the difference in the number of pixels between the original image and the reproduced image is determined by the control unit 4.
Controlled by 00. For example, when linear density conversion is performed at a conversion rate r/R, the above multi-value data calculation is performed once every R times of the reference clock output from inside the control unit 400, and the input of original image data is performed 91 times per clock. One group rushes to However, in order to match the starting points of the original image and the reproduced image, the first operation of both processes is executed at the first clock. As a result, from the original image of X pixels in the main scanning direction and 7 pixels in the sub-scanning direction, X*r/R pixels in the main scanning direction, T
iA scanning line direction Y*r/R multi-value data of the image device is obtained, and by binarizing the multi-value data, main scanning line direction X*r/R pixels and sub-scanning line direction Y*
A reconstructed image of r/R pixels is obtained. At this time again,
The pixel at the coordinate (xt y) on the reproduced image corresponds to the coordinate (x*R/r, y*R/r) of the original image. Note that from now on, the binary data of the coordinates (XxyYt) of the original image will be expressed as P (Xte
yt), and the binary data at the coordinates (x2°y2) on the reproduced image is expressed as Q (X 21 '/ +1).

Next, each part of the apparatus will be described in detail using linear density conversion processing as an example. First, the operation of obtaining multivalued data S(x,y) from binary original image data P(xwy) will be explained.

FIG. 9 is a block diagram of the processing section for obtaining multi-value data S (x, y) using this method. corresponds to The number of pixels to be referred to when determining multivalued data is arbitrary, but here, a case will be explained using an example in which binary data of 4*4 pixels is referred to.

Here, the conversion rate of linear density conversion is expressed as r in the X direction, / R1g
The coordinates of the reproduced image (x2
The coordinates (Xx*'/l) of the center pixel of the original image data are referred to in order to obtain the data of ゜yz).

x 1 = [x 2 * Rt / r 1] y 1 =
Defined as Cy2*R2/r21.

However, in normal processing, the conversion rates r 1/R1 and r*/R2 often become r 1/R1=r 2 /R2.

When calculating the multivalued data s (x2t y2), the binary data to be referred to is the coordinates (
This is 166 pixel data including X1#3'l). now,
Original image data P (X1+2*V
1+2), the latch column 121. The three latches of P (Xx+t* yx+2L
P (Xte yt+2)yP (Xz-to Vt+
z) is recorded, and from the four output lines 131, the binary data for the three pixels and the original image data P (x 1 + * ey t -1-2) input at that time are recorded.
Power output. Similarly, the binary data for four pixels is transmitted from the two signal lines 132, 133, and 134 to the multi-value data determining unit 2.
00, and from a total of 16 signal lines, I! (”
! -1tVx-t) to (xt+at yt+z)
166 pixel data are output respectively. This one
As shown in FIG. 6, the 66-pixel data is multivalued data S (
This is the data for determining X21 yz).

Also, ΔX and Δy in the figure are Δx=x2*Rx/r, -x.

ΔV=312*R11/rx−Yt. Therefore, the data output from the signal line 130 is input to the multi-value data determining section 200. The multivalued data determination unit calculates S (x 2t y s3) using the 166-pixel binary data input from the signal line 130 and the weighting coefficient α (ΔX, Δy) for 4*4 pixels determined by ΔX and Δy.
This is the part that determines and outputs. Here, the 16 binary data inputted from the signal a 130 are multiplied by a weighting coefficient α (ΔX, Δy) for each pixel by an integrator 210, where the weighting coefficient is determined by ΔX and Δy.
It is determined by the coefficient determination unit 230. An example of α is the 10th
As shown in the figure. Further, ΔX and Δy are counted by a counter inside the control unit 400, and
Input by In addition.

Instead of ΔX and Δy, x4=mod (x2*R1, r 1)y 4=mod
(y 2 * R2, r 2) X4# y4 is r
Even if input with l*r11, Δx4xa/rt ΔY = y 4 / r 2 The same processing can be executed from the second. In addition, the coefficient determination unit 23
0 can usually be realized by ROM. The data are added by an adder 220 and output as multivalued data 5 (x2ty*). Further, if the signal line 130 and the signal lines 241 and 242 are used as address lines, the entire multilevel data determining section 200 can be realized with one ROM. A block diagram in this case is shown in FIG. If all 166 pixel data lines are used as address lines, the ROM required for the multi-level data determining section 200 will require an extremely large capacity.

Therefore, it is effective to reduce the number of address lines by forming groups of 166 pixels, for example, several groups. 250 in FIG. 11 indicates the converter. In addition, 240 in the figure indicates the output of the converter 250 and the signal lines 241 and 2.
ROM (
This can also be achieved using Read Only Memory).

Next, the operation of the control section 400 will be explained using FIG. 12. In FIG. 12, 410 is a clock that outputs a reference pulse, and 420 and 430 are address counters that count addresses Xxe'12 of output reproduced image data. On the other hand, the address X 1 e Y 1 of the input original image data is stored in the address counter 44.
0 and 450 respectively. We will now explain the case where the conversion is performed using □/Rt in the X direction and r2/R2 in the y direction.In this case, the input of the original image data is carried out once per reference pulse r, and is converted multiple times. The value data is determined once every R1 times. Therefore, the address counters 420 and 440 receive the reference pulses rl and 440, respectively.
Make one change every R□ times.

Also, each time you input original image data for one scanning line,
It is necessary to advance the address of the reproduced image in the sub-scanning line direction by R2/r2. Therefore, when the address counter 420 finishes processing one scanning line, the end detection section 415 outputs the pulse signal r times. Hereinafter, the pulse signal output from the end detection section 415 will be referred to as a line pulse. Address counters 430 and 450 each have a line pulse r
One change operation is performed every two times of 2tR. However, multivalued data s (
X2172), yK image data P
(Xx+2*)'I+2) must already be input. Therefore, it is necessary to input the original image data ahead of the output of the address counter 420 by a number of pixels and ahead of the output of the address counter 430 by two lines. In addition, in order to perform processing at an arbitrary conversion rate, rl
, R1, r2°R2 must all be able to be instructed from the outside. However, in actual use.

There is no problem in setting rivr2 to a relatively large fixed value. Therefore, in this embodiment, a case where only R1 and R are inputted from the outside will be described as an example.

Counters 460 and 470 increment their counts by reference pulses and line pulses, respectively, and r1*, respectively.
If the output is returned to 0 with r2 pulse inputs and the resulting outputs of counters 460 and 470 are set to X4#14, then ΔX and Δy are obtained as ΔX = , rl and r2
If is a constant, then x4 and y4 are directly connected to the signal line 24.
1 and 242 as an address of the coefficient determining section 230 or the ROM 240 for determining multivalued data, the weighting coefficient α (ΔX, Δy) can be determined. In this example method, binary data for one pixel is obtained every time R1 pulse signals are output. Therefore, high-speed processing is difficult as it is. In processing one image, the conversion rate r1. r2. R□.

Since R2 is usually unchanged, the address of the reproduced image Xx*
Addresses of the original image XzeVz and Δ, x for V1
, Δy change periodically. Therefore, at the beginning of processing
2y312, ΔX, Δy, or XzeVx and α(Δ
This problem can be solved by determining X, Δy), storing it in a line buffer, for example, and outputting it periodically. An example using this method is shown in FIG. In this system, the device has two modes: a Q-equipment mode for determining X2172, ΔX, and Δy, and an image processing mode for actually performing image processing. When the process starts, first, a signal is sent from the signal line 401 indicating that it is in the preparation mode. Therefore, the clock 410 outputs the reference clock for one scanning line, and the address counter 440 outputs the reference clock for one scanning line.
and records the output of counter 460 in line buffers 445 and 465. Meanwhile, at the same time, the line pulse generator 411 outputs a number of line pulses corresponding to the number of scanning lines of the image, and the outputs of the address counter 450 and the counter 470 are recorded in the line buffers 455 and 475 for each line pulse R2. Line pulse generator 4
11 operates only in the preparation stage, and the clock 410 can also be used. Thereafter, actual image processing begins and a signal indicating that the mode is 1 image processing mode is output from the signal line 401. Thereafter, the entire system is controlled only by the reference clock output from the clock 410, and the address counter 420 operates by inputting clock pulses less than R1 times. Therefore, faster processing can be performed without changing the frequency of the reference clock. Then, at the same time as the output of the address counter 420 changes,
Line buffers 445, 455, 475 correspond to X2172. ΔX and Δy are read out and sent to the multi-value data determining section 230. Here, line buffer 445
, 455 are controlled by an address counter 420. On the other hand, line buffer 4
The address of the output data of 65, 475 is controlled by the address counter 430.
0 operates every time the address counter 420 operates for one scanning line in the image processing mode. Also, 425 in the diagram
is a selector that selects either the output of the address counter 420 or the output of the line pulse generator 411 as a line pulse, and is controlled by a signal on the signal line 401.

Next, the operation of the binary data determining section will be explained in detail. This part is done by converting the multivalued data into binarized data.

This is the place where the process to obtain binary reproduced image data (process (2) described in the [Operation] section) is executed.Many methods can be applied to this configuration, including the conventional binarization processing method. Here, we will discuss in detail the case of using the two types of methods described in the [Operation] section.The first method is
The example shown in FIG. 14 will be explained. Multivalued data S (x
, y) are input to the adder 330 through the signal line 301.

The adder 330 adds S(x, y) and multi-value data E (xt y) calculated by a method described later, and outputs multi-value data F (X*y). Here, multivalued data E(x,
y) is input via the signal line 331. F(x+'y)
The signal 13311 is input to the comparator 310,
It is compared with a predetermined threshold T. Based on the result of this comparison, the binary data Q (xt y) of the reproduced image
is determined.

Note that the multi-value data E (x, y) used here is obtained by 2
An example of the 0 weighting coefficient β, which is the sum of the error ε generated when converting into a value, by a predetermined weighting coefficient β, is shown in the 15th example.
As shown in the figure. In the figure, * indicates a pixel to be binarized at that point, which corresponds to the pixel at the coordinates (x, y) mentioned here. In this case E (x.

y) is obtained by the following formula.

E(x, y) = 1/8 (i(x 1.y-1)+
t(x+1.y-1)+ε(x-2,y)+ε(x,y
-2)+2(i(x,y-1)+1(x-1,y)))
On the other hand, the error ε of each pixel is either the difference between the multivalued data F and O, or the difference between F and the maximum value max that F can take. This value is obtained as follows. At a certain point, when the comparator 330 binarizes the multivalued data F(X2-1eyz) at the coordinates (X2-1.y2), F(X2
-t+Yz) is input to the comparator 310, as well as the differencer 335 and selector 340. The differentiator 335 calculates the maximum value max and F of the values of the multivalued data S (x, y).
Output the difference of (X2 +lt yz) and selector 390
send to For example, in this example, S (x, y) takes values from 0 to 4, so max=4, and the difference unit 335
Output S (X2-1*y2). Also, here, if F (x 2 - x y y 2) > max, the differentiator 335 outputs O. The selector 340 is.

F (x2-11 y2) or m according to the following conditions
ax-F (X2-1e y2) with error ε
Output as (Xu-x+3!t).

t(x, y)=F(x, y): Q(x,
y)=Omax-F(x,y): Q(x,y)=1
The output t(x2-□, y2) is sent to the line buffer 350 through the signal line 351. −E (xt y
) is executed by the latches 346, 347, 348 and shift registers 371, 372.

The operation of each part will be explained using the case of finding y) as an example. 2
When the value data Q (x l * y) is determined by the binarization processing unit, the latches 346, 347° 348 and the output lines 351, 352, 353 of the line buffer 350, 355
, 354, 355°356, the error is 1(x
l+y), i (x-2*y), ε (Key-
1), i (x-1, y-1), s (x+1
．． y-1).

t(x+y2) is output. Here, the signal line 35
1 and 354 data & (X-1t y), ε (x
, yl) are entered into shift registers 371 and 372, and 2
*t (x-1, y), and 2eg (x, y
1) is output. The adder 380 adds the input six types of multi-value data to calculate 8*E (xp y), and sends it to the shift register 373. shift register 37
3, by shifting the input multi-value data, E
Obtain (xt y). The obtained E(xyy) is input to the adder 330 via the signal line 331. In addition, xp1
, or if y=l, it happens that the part necessary to obtain E (xt y) does not exist. in this case,
E (*, y) is determined by drawing the values recorded in advance in line buffers 350 and 355.

According to this method, since the threshold value at the time of binarization can be fixed, it is possible to obtain a high quality reduced image without moiré even for an image with a high periodicity.

Next, the second method will be explained. FIG. 16 is a block diagram of the above method. Multivalued data S (x.

y) is a first multi-value data storage section 510 and a second multi-value data storage section 5, each having a capacity for two scanning lines, located in two locations.
15 is temporarily saved. This multivalued data first storage section is
For example, it is configured by a combination of line buffers and latches. Furthermore, the line buffer variables and the number of latch stages are determined by the number of pixels referenced in -degrees. Although the number of pixels to be referenced is arbitrary, in this embodiment, 2e pixels will be explained as an example. Of the two storage sections, 515 stores input multivalued data. On the other hand, 510 replaces internal data one after another. Hereinafter, the result of mo substitution on input data S will be written as Sm. If 5 (x5.ys) is input from the signal line 201, the coordinates (xs-he Ys-IL (Xse
7s-IL (Xs-xe ysL (Xis ys
) are multivalued data s, (XS-1e Ys-
ILS2 (x6w Ys-tL St (Xs-tt
Vs)eS(Xspys) is stored. Therefore, multivalued data Ss 0Cs-xe Ys-thes2 (xs
t Ys-IL sl (Xs-1y ysLS
(xst7s) and sends it to adder 520.

The adder 520 connects these four multi-value data and the signal line 590.
The input error correction amount E is added to the multivalued data Sm.
(Xs-1e Ys-t) is output.

The multivalued data Sm (xs-xe ys-t) is then used in a divider 530 to obtain integers N and n using the following equations.

Sm (Xs-1, Ys-t)=N*M+n where 1M
is a predetermined constant, and usually uses the maximum value that multivalued data S(x, y) can take. - The signal lines 521 and 52 are also connected to the line buffer in the power cylinder 2 storage unit 515.
Multi-value data S (XS1-Lt
Ys-ILS (Xse ys-IL S (xs-
xt ysLS (xstys) is sent to the descending order determiner 540. The descending order determiner 540 orders the input four types of multivalued data in descending order of density. The determination result of the descending order determiner 540 and the outputs N and n of the remover 530 are input to the replacement unit 550. The replacement unit 550 assigns N pixels to M to the determiner of the descending order determiner 540, gives n to the N+1st pixel, and gives 0 to the other pixels, resulting in multivalued data S 3 (X
5-1 t V 6- ILs2 (xst Ys-IL
sl (XS-1s ysLS(x5.y5) are respectively S 4 (X 6-1* V, 6-1)tS3 (X
stys-tL S2 (XS-Xs ys), tSi
(xst ys). Note that Si (xs
tysL 52 (Xs-1s ysLS3 (xst
ys-t) is recorded in the first storage unit 510 because the replacement process cannot be performed again while the process continues. but.

S4 (X5-1+'/5-1) is not recorded in the first storage unit 510 and is stored on the signal line 55 because the pixel at the coordinate (X6-1ey5-1) is not referenced again.
1 to the comparator 560. Comparator 560 is S4
(XS-1+y5-x) is compared with a predetermined threshold T, and binary data Q(XS, -□) is obtained.

y6-t) is determined. In this comparison, S 4
When (X 5-1 + 16-1) is n, an error occurs due to binarization. Therefore, similar to the method explained earlier,
This error will be referenced during subsequent processing.

S4 (x6-1+)'5-1) is input to the comparator 560, as well as to the differentiator 570 and selector 580 at the same time.

The differentiator 570 calculates S 4 (X S -1+ 15-1
) and T (X5-LtV5-1) is output and sent to the selector 580. The selector 580 uses the binary data Q (xs-1w ys-z) to select S4 (XS-t* ys-t) and G (x6-
1°'15-t) and outputs it as the error correction amount E.

E-84 (xt y): Q (xt y) = O
G (xt y): Q (xt y) = 1 E obtained here is the sum of multivalued data for determining the next pixel Q (XspYs-t), Sm (Xst ys-x'). This is the error correction amount added at the time of determination.

As explained above, in the present invention, various methods can be applied for binarization processing of multivalued data, so the degree of freedom in device configuration is large, and a processing device suitable for individual output devices such as CRTs and printers can be realized.

With the apparatus described above, high quality reproduced images can be obtained with a relatively simple configuration. .

〔Effect of the invention〕

According to the present invention, since multivalued data is determined from a plurality of binary data, pixel data at positions corresponding between pixels of the original image can also be determined without missing data at specific coordinates. Further, since the binarization process of multivalued data can be executed using a fixed threshold value, deterioration of image resolution can be prevented. As a result, image interpolation processing can be performed with high precision, allowing for image enlargement.

When processing such as reduction and nine rotations is performed, a reproduced image of high image quality can be obtained.

[Brief explanation of the drawing]

FIG. 1 is a block diagram showing the basic configuration of the device. Figure 2 is a diagram explaining the relationship between the original image and the reproduced image by linear density conversion, Figure 3 is a diagram explaining the positional relationship between the pixels of the original image and the reproduced image, and Figure 4 is the conventional thinning process. FIG. 5 is a diagram illustrating the enlargement processing method for conventional example 1, systematic dithered image, and FIG. A diagram for explaining the positional relationship of the original image data to be referred to, FIG. 7 is a diagram for explaining the principle of an example of the binarization processing method for multivalued data used in the present invention, and FIG. Figure 9 is a diagram illustrating the principle of an example of the binarization processing method for multivalued data used in
Figure 10 shows an example of a weighting coefficient matrix used for determining multi-value data. FIG. 12 is a diagram showing an example of a control unit realized with a simpler configuration. FIG. 13 is a diagram showing an example of a control unit for high-speed processing. Figure 15 shows an example of the configuration of a binarization processing unit, Figure 15 shows an example of a weighting coefficient for determining the amount of error correction, and Figure 16 shows an example of the configuration of a binarization processing unit. The figure which shows one example of a structure. [Explanation of symbols] 100... Image data storage section, 101... Original image data input line, 111... Line buffer, 112
...Line buffer, 113...Line buffer,
121...Latch row, 122...Latch row. 123...Latch row, 124...Latch row, 130
...Signal line, 200...Multi-value data determination section. 210... Multiplier, 220... Adder, 230...
- Weighting coefficient determination unit, 210... Multiplier, 220...
Adder, 230...Weighting coefficient determination unit, 221...Output line, 240...Multi-value data determiner, 241...Signal line, 242...Signal line, 250...Converter, 30
0...Binarization processing unit, 310...Comparator, 320...
...Threshold value determination unit, 330... Adder, 335... Differentiator, 340... Selector, 346... Latch, 3
47...Latch, 348...Latch, 350...
line buffer. 360... Line buffer, 371... Shift register, 372... Shift register, 373... Shift register, 380... Adder, 391... Reproduction image data output line, 400... Control 401... Signal line, 410... Clock, 411... Line pulse generator, 415... Pulse generator, 420... Address counter, 430... Address counter, 44
0...address counter, 450...address counter, 460...counter, 480...counter,
425...Selector, 445...Line buffer. 455... Line buffer, 465... Line buffer, 475... Line buffer, 510... Original image data first storage section, 515... Original image data second storage section, 520... Addition device, 530... divider, 54
0... Descending order determiner, 550... Replacement unit, 560...
-Comparator, 570...Differentiator, 580...Selector. (a) tb) Paper S round (, +23 15 figure CC) td) Kaya s1! ! (b) Figure 7θ Akira/Figure 1 S51): y? a% 4to Rimuri 4M -7) "Jibi Kushita
42r Yare 7 4/l Raipa 1 Yushi 1 Su6
σ Gakunta φ4 4 Konoifuku 1420
71? nisu gau〉ta ηunta 4t
! ; LaAnanzufufa Figure No. 5

Claims (1)

[Scope of Claims] An image processing device that performs image processing on one- and two-value digital images, including an image storage unit that temporarily stores binary data P of an original image, and an image in the image storage unit. M data
means for scanning using a scanning window of _1 pixel (M_1>1); means for determining multi-value data S from binary image data P of M_1 pixels in the scanning window; An image processing device characterized by having means for converting the image into an image. 2. In the image processing apparatus according to claim 1, the means for binarizing the multivalued data S includes an error correction amount E.
means for adding the multi-value data S to the multi-value data F to obtain the multi-value data F;
It has means for comparing the sum F of the multivalued data with a predetermined value C, and outputting binary data Q such that if F>C then Q=1 and if F<C then Q=0, and the error correction amount E An image processing apparatus characterized in that a value obtained by multiplying a rounding error ε generated when neighboring pixels are binarized by a certain coefficient is added over a certain range. 3. In the image processing device according to claim 2 above,
The means for determining the binary data Q from the multi-value data S is as follows:
a first storage section and a second storage section for storing the multi-valued data S; and means for scanning the first storage section with a first scanning window of M_2 pixels to obtain a sum S_m of the multi-valued data S in the scanning window. , a means for determining the sum F of the S_m and the error correction amount E, an arithmetic means for determining an integer N such that F=C*N+n from the multi-value data F and a predetermined value C, and the first memory. A second scanning window scans the second storage unit in synchronization with the scanning window in the second scanning window, and a specific correction value is added to the multi-value data for M_2 pixels in the second scanning window, and the numbers are numbered in descending or ascending order. and according to the number, the data for N pixels in the first scanning window in the first storage section is stored in the C, the multi-value data of the pixel with the number N+1 is stored as the n, and the multi-value data of the remaining pixels are stored in the C. The means for replacing with 0 and the operations of the above series of means are repeated to obtain M for each pixel.
Multi-valued data S of a pixel in the first storage unit that corresponds to a pixel for which replacement processing has been performed twice and has already been replaced M_2 times
Due to _x, if S_x≧V then Q=1, E_1=C−S_xS_x<
An image processing apparatus comprising binary data Q such that Q=0 and E_1=S_x if V, and means for determining an error correction amount E_1 to be added when binarizing the next pixel.