JPH0420224B2 - - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION The present invention relates to a method for enlarging and reducing digital images, and particularly to a method suitable for rapidly obtaining high-quality enlarged and reduced images at arbitrary magnifications.

The principle of enlarging and reducing digital images will be explained with reference to FIG. In the figure, a is an example of the original image, and b is an example of the enlarged/reduced image.
The image is shown enlarged 3/2 times horizontally. Figure c is
The enlarged image b is mapped onto the original image a. Enlarging/reducing processing means obtaining an enlarged/reduced image b from the original image a, which means that in the figure c, the values of the grid points q ₁₁ , q ₁₂ , ..., q ₂₁ , ... of b are
This results in interpolation based on the values of grid points P ₁₁ , P ₁₂ , ..., P ₂₁ , ... of a.

Conventionally, various methods have been proposed for the interpolation rules. For example, SPC method, logical sum method, 9 division method,
Projection methods and the like are well known. (Refer to the proceedings of the 20th National Conference of the Information Processing Society of Japan in 1978, pp. 463-464, etc.). Although the interpolation rules themselves are different for each of the above methods, the input information for the interpolation calculation is common to each method. This input information will be explained in detail below with reference to FIG. q _{n , o} is _the interpolation point _{, P} Vertical address, A is the horizontal length of the original image grid,
B is the same vertical length, a is the horizontal length of the enlarged/reduced image grid, b
have the same vertical length. Among these, A and B are constants unique to the device, and a and b are constants uniquely determined by the magnification. The input information required to calculate the value of the interpolation point q _n, o is x _o , y _n , four nearby points, P _{(Xo, Yn)} ,
These are the values of P _(Xo+1,Yn) , P _(Xo+1,Yn+1) , and P _(Xo,Yn+1) .

Values in Figure 2, x _o , y _n , A, B, a, b
is a discrete quantity. This is because there is no problem in practical use even if the scaling factor is limited to rational numbers. Since the following is simpler than this precondition, the above x _o , y _n , A,
It is assumed that B, a, and b take values converted into integers.

In the conventional enlarging/reducing apparatus, the input information necessary for the interpolation described above is obtained by the procedure shown in FIG.
(For example, see the proceedings of the 22nd National Conference of the Information Processing Society of Japan in 1981, pp.73-74).

It is defined as U _o =X _o・A+x _o V _n =Y _n・B+y _n . In the figure, at 301, V ₀ = 0, at 302, y _n = V _n mod B, at 303, Y _n = V _n /B, at 304, U ₀ = 0, at 305, x _o = U _o mod A, at 306, X _o = U _o /A 307 stores the values of four neighboring points P _(Xo,Yn) , P _(Xo+1,Yn) , P _(Xo+1,Yn+1) , P _(Xo,Yn+1) in the image memory. Load more. In 308, the above 30
Interpolation calculations are performed using the results of 2, 305, and 307 as input information. In 309, U _o+1 = U _o +a n=n+1, and in 311, V _n+1 =V _n +b m=m+1. However, in the above procedure, mod is an operation for obtaining the remainder of division, and / is an operation for obtaining the quotient of division.

It has been confirmed that if the above procedure is followed, it takes about 1 second to obtain a scaled image consisting of 10 ⁶ points. This corresponds to the fact that it takes about 4 seconds to obtain an enlarged/reduced image corresponding to the size of an A4 paper using a facsimile machine with a scanning line density of 8 lines/mm. In the above steps,
Actually, 305, 306 and 307, 308 can be calculated in parallel. At this time, it is the passes of processes 306 and 307 that govern the processing time.

An object of the present invention is to speed up the scaling processing by various conventional methods by speeding up the passes of the processing 306 and 307.

This invention provides the above-mentioned sequences x _o , y _n and x _o+1 , −
The above object is achieved by utilizing the fact that the sequence X _o , Y _n+1 -Y _n is a periodic sequence that repeats the basic sequence at a fixed length period. Specifically, this is achieved by initially loading each of the basic columns into a rotating shift register, sequentially rotating and shifting this using a timing pulse, and extracting information.

First, the periodicity will be explained. First, let me explain about x _o . x _o can be defined as x _o = U _o mod A U _{o +1} = U _o + a as described above. At this time, for any n, x _{o + A} = U _{o + A} mod A = (U _o + A · a) mod A = U _o mod A = X _o holds true. Therefore, x _o is a periodic sequence with period A at most. The basic sequences are x ₀ , x ₁ , ..., x _A-1 .
In exactly the same way, it can be proven that y _n is a periodic sequence with a period of at most B, and the fundamental sequences are y ₀ , y ₁ , . . . y _B-1 . Next, X _o+1 −X _o will be explained. As mentioned above, X _o can be defined as A·X _o = U _o −x _o . Therefore, A・(X _o+1 −X _o )=a−(X _o+1 −X _o ). At this time, if we set X _o+1 −X _o =ΔX _o , then for any n, A・ΔX _o+A = a−(x _o+A+1 −x _o+A ) = a−(x _{o +1} −x _o ) =A・ΔX _o holds true. Therefore, ΔX _o+A = ΔX _o holds true. Therefore, ΔX _o =X _o+1 −X _o is a periodic sequence with a period of at most A. The basic columns are ΔX ₀ , ΔX ₁ ,
..., ΔX _A-1 . In exactly the same way, it can be proven that ΔY _o =Y _o+1 −Y _o is a periodic sequence with a period of at most B, and the basic sequences are ΔY ₀ , ΔY ₁ , . . . , ΔY _B-1 .

Next, one embodiment of the present invention will be described in detail. FIG. 4 shows the overall configuration of the present invention. 401
402 is a processing device such as a microcomputer, and 402 is a parallel-to-serial converter (hereinafter abbreviated as P/S converter) that converts original image data read out in parallel format from the main storage device in the processing device 401 into serial format; 403 is a buffer memory 404
a demultiplexer for selecting the input line of 40;
Reference numeral 4 denotes a buffer memory using a shift register that stores original image data line by line for a specific number of lines (for example, 4 lines); 405 denotes a buffer memory 4;
Multiplexer to select output line of 04, 4
06 inputs the values of neighboring pixels of the interpolation point from the multiplexer 405, and also inputs the in-lattice addresses x,
An interpolation calculator 407 inputs y and outputs the value of one pixel of the enlarged/reduced image, and an interpolation calculator 407 converts the enlarged/reduced image data output in serial format from the interpolation calculator 406 into a parallel format. A serial-to-parallel converter (hereinafter abbreviated as S/P converter) that writes to the storage device, 408 is a processing device 401
A read/write controller (hereinafter abbreviated as R/W controller) 409 controls the transfer from the buffer memory 40 to the P/S converter 402 and from the S/P converter 407 to the processing device 401.
Rotating shift register 41 for the number of shifts of 4
1, and a shift controller that outputs this number of shift pulses; 410, a rotating shift register that sequentially outputs the address information of the main memory in the processing device 401 to be taken into the buffer memory 404 every time one line is interpolated; 411, a buffer memory; a rotating shift register that sequentially outputs 404 shifts each time one pixel is interpolated;
412 is a frequency divider that counts the interpolation of one pixel for one line, and 413 is the in-grid address of the interpolation point.
A rotating shift register 414 sequentially outputs the coordinates each time one pixel is interpolated, and 414 is a rotating shift register that sequentially outputs the y coordinate of the in-grid address of the interpolation point each time one line is interpolated. Correlating this diagram with the symbols in Figure 2, x _o is the output of the rotating shift register 413, y _o is the output of the rotating shift register 414, and X _{o +1} - X _o = ΔX _o is the output of the rotating shift register 411. , Y _n+1 - _{Y n} = ΔY _n is the output of the rotating shift register 410, one pulse every time n counts up is input to the registers 411 and 413, and one pulse every time m counts up is input to the register 410. and 414 input,
Px _o+1 , Y _n and Px _o+1 , Y _n+1 are multiplexers 40
The output will be 5. In particular, the neighboring pixels of the interpolation point are ΔX _o
In order to obtain the buffer memory 404 using
Use shift registers.

Next, the operation of this embodiment will be explained with reference to FIG. 501 is the scaling factor, horizontal direction A/a,
Vertical direction B/b (However, A=B is device specific)
Based on this, the processing unit calculates the basic columns of x, ΔX, y, ΔY, and calculates the basic columns of x, ΔX, y, ΔY, and
Load to 0,411,413,414. 50
2, the first two lines of the original image are transferred to buffer memory 4.
Load into 04. In 503, based on ΔY,
Calculate the addresses of the next two lines of the original image. 504
Now, based on the address 503, the next two lines are loaded into an empty line in the buffer memory 404.

In steps 505 and 507, the contents of the two shift registers being processed in the buffer memory are shifted by one bit, and the two overflowing pixels are taken into the interpolation calculator 406. In 506, it is determined that the number of shifts in 507 becomes ΔX _o times. In 508, x rotates and shifts ΔX in 509 to obtain x _o and ΔX _o , respectively. At 510, interpolation calculations are performed based on x _o , y _o , and the latest four pixels overflowing from the buffer memory. In step 511, it is determined whether interpolation for one line has ended. By rotating and shifting y in 512 and ΔY in 513, y _n+1 and ΔY _n are obtained, respectively. In 514, it is determined whether interpolation for one image has ended.

In this embodiment, in the process 50 in FIG.
6 and 507, 508, 509 and processing 512,
513 and processing 503 and 504, 505-51
3 performs parallel processing. Among these, 505-51
The total processing time of the third pass is equal to the shift time of the number of pixels of the enlarged/reduced image in the shift register. this is,
It takes about 0.6 to obtain an A4 scale image with a pixel density of 8 lines/mm using A _n 2827 as a shift register.
This corresponds to taking seconds. the other 503 and 504
The total processing time for the pass is equal to the total number of bytes in the original and scaled images multiplied by the bus transmission speed.
This corresponds to the fact that it takes about 1 second to obtain the A4 size image using a 1 Mbyte/sec bus.

As explained above, according to the present invention, FIG.
By replacing the process of 7,309 with steps 505 to 509 in FIG. 5, the A4 process that conventionally took about 4 seconds
This has the advantage that the printing image can be enlarged or reduced in about 1 second.

[Brief explanation of drawings]

1a, b, c and 2 are explanatory diagrams of image enlargement/reduction, FIG. 3 is a flowchart showing the processing procedure of the conventional method, and FIG. 4 is a block diagram of an embodiment of the present invention. FIG. 5 is a flowchart showing the processing procedure according to the present invention.

Claims (1)

[Claims] 1. From an original image consisting of a plurality of pixels arranged in a grid in the row and column directions, a plurality of neighboring pixels are selected corresponding to each grid point on a converted image with a specified magnification, and these pixels are In an image scaling method that determines the pixel value of each grid point on the converted image based on the value of neighboring pixels of the converted image, a plurality of neighboring pixels from the original image are An image enlarging/reducing method characterized in that pixel selection is performed, at least in the row direction, using predetermined periodicity data determined by the specified magnification. 2 Periodicity data regarding the positional relationship between each grid point on the converted image determined according to the specified magnification and a plurality of neighboring pixels on the original image is stored in advance, and each grid point on the converted image is stored in advance. 2. The image scaling method according to claim 1, wherein the pixel value of is determined based on the pixel value of each neighboring pixel selected from the original image and periodicity data regarding the positional relationship.