JPS6097474A - Method and apparatus for rotating picture - Google Patents

Abstract

PURPOSE:To obtain a rotation picture at high speed in digital picture processing by making oblique axis conversion and enlarging and reduction in one direction by update from original picture data to coordinate data and further making enlarging and reduction and oblique axis conversion in another direction. CONSTITUTION:Affine conversion that indicates rotation can be resolved into the formula 5, and the conversion can be attained by executing successively conversion of the vertical y direction oblique axis conversion formula 6, vertical direction reduction conversion formula 7, horizontal x direction enlarging conversion formula 8 and horizontal direction oblique axis conversion formula 8. These processes are performed on a digital picture of a picture memory 105 inputted from a picture inputting device 103 by a microprocessor 101, a program memory 102, a processing section 101 etc. Conversion 6 is made on integer values m, n, and picture element density of coordinates m1, n1 of the converted picture is made picture element density of coordinates m, n of the converted picture. Conversion 7 is made on the integer values m1, n1, and a rotation picture is obtained by making conversion 8, 9.

Description

[Detailed Description of the Invention] [Field of Application of the Invention] The present invention applies to digital image processing. [Background of the Invention] Rotation is one of the basic processes in the field of image processing.
For example, correcting the skew of an input document in facsimile, superimposing seal impressions in seal verification (for example, ``Seal imprint pattern alignment method using correlation of peripheral density with respect to centroid'', author Toru Kaneko et al., Institute of Electronics and Communication Engineers, IB81-
13. This is necessary for systems such as pages 37 to 42).

Conventionally, typical methods of rotation include a method using affine transformation and a method using oblique axis transformation. The method using affine transformation is as follows: Equation transformation (1) is the definition of rotational transformation in geometry itself, and does not cause distortion in the rotated image, but has the drawback that it is difficult to speed up transformation (1). The method using oblique axis transformation approximates the affine transformation (1) of rotation by performing oblique axis transformation twice in the vertical and horizontal directions. 95145, inventor Hitoshi Miyai) etc. Specifically, transformation (1) is approximated by longitudinal oblique axis transformation and twist. Oblique axis transformation (2) and (3
) allows faster processing than conversion (1). Therefore, this rotation method is faster than a method using affine transformation. However, as the rotation angle θ increases, the rotated image becomes distorted (for example, at θ-1, a square becomes a parallelogram with one vertex angle of 1), and other drawbacks often occur.

[Purpose of the invention]

The purpose of the present invention is to eliminate the drawbacks of the conventional method as described above,
An object of the present invention is to provide an image rotation method that is capable of high-speed processing and does not cause distortion in rotated images.

[Summary of the invention]

In order to achieve the above object, the present invention has been made based on the following idea.

Normally, when storing images in memory for various processing,
Pixels that are continuous in the horizontal or vertical direction on the image are made to correspond to areas that have continuous addresses on the memory.

Particularly in the case of binary images, pixels that are continuous in the horizontal or vertical direction can usually be read and written as one word. The reason why it is difficult to increase the speed of the conventional method using affine transformation is that even if the original image is read out in the order of pixels with consecutive addresses in the memory in the transformation (1), the converted image is generally one pixel in the memory. This is because each address must be written to a memory area that is not consecutive. On the other hand, the method using the oblique axis transformation can be speeded up by the transformation (2)
This is because in (3), the continuous area of the original image corresponds to the continuous area of the converted image, excluding some image boundaries. Therefore, especially when dealing with a binary image in which one word has information of multiple pixels, affine transformation is inefficient as it can only process one pixel unit, but oblique axis transformation can only process one word at a time. It is possible and efficient. In addition to the above-mentioned oblique Il'l]1 transformation, conversion methods that are lazy and can be processed at high speed include horizontal or vertical scaling conversion.

The rest of the present invention is based on the fact that affine transformation (1) can be expressed by a combination of the above-mentioned oblique axis transformation and the above-mentioned scaling transformation, and transformation (1) is executed by sequentially executing these transformations. At the point. Since the combined transformation is equivalent to affine transformation (1), no distortion occurs in the transformed image. Each transformation used for silk matching can be processed at high speed as described above, and as a result, speeding up of rotation processing can be expected.

[Embodiments of the invention]

Hereinafter, the present invention will be explained in detail using examples.

Regarding the affine transformation (1) representing rotation, the decomposition given as an example holds true. (However, θ≦1) Therefore, transformation (1) sequentially executes four transformations: vertical (y) direction oblique axis conversion, vertical direction reduction, horizontal (X) direction enlargement, horizontal direction oblique axis conversion This can be achieved by doing so. There are other sets of oblique axis transformation and scaling transformation that are equivalent to transformation (1), but the minimum number of transformations is four as shown above. The outline of conversions (2 to (9) is shown in FIG. 1.

However, in the digital image targeted by the present invention, the density is defined only on the sampled coordinates, so interpolation is required to actually execute transformations (6) to (?2). In this case, the nearest neighbor selection method (for example, [Image Rotation Simulation Using II Calculator], author Kageyama, 1985 National Conference of the Institute of Electronics and Communication Engineers, A1067) is used as the interpolation method. is realized by the following steps.In the following explanation, the unit of coordinates is taken as the interval between the sample points mentioned above.

[Step 1] Transform (6) is applied to the integer values In, H, and xl, m
, nYr, □, n are obtained. In other words, the following coordinate transformation is performed.

Next, Xl, m, n*Y1+lTl+rl are rounded off to integers, and the results are set as ml and nl, respectively (interpolation process).

Transformation (7) is applied to the integer values ml and nl, and X2.
We obtain ml, nl + Y 2 , ml, n 1.

In other words, the following coordinate transformation is performed.

Next, x2. ml, nl T Y 2. ml, nl are rounded off to an integer value m2. Transformation (J') is applied to A2, and xl, A2. Re 2 r Y 8. A2. Get A2. In other words, the following coordinate transformation is performed.

Next, xl, A2. A2 + Y 3. A2. A2
[Step 4] Integer value m3. Apply conversion ('?) to A3 and convert it to X A
3 A3 + Ym3 A3 is obtained. In other words, the following coordinate transformation is performed.

Next, X A3. A3 + Yrn3. FIG. 2 shows the pixel correspondence of the step VC, h, and r de Sakaki shogyo images when A3 is rounded off. 1 to 10 in the diagram
is a pixel number, and pixels with the same number are corresponding pixels.

In steps 1 and 3 above, a set of pixels that are continuous in the vertical direction on the converted image appears as is as a set of pixels that are continuous in the vertical direction on the converted image. Furthermore, in Steps 2 and 4, a set of pixels that are continuous in the horizontal direction on the converted image appears as is as a set of pixels that are continuous in the horizontal direction on the converted image. Therefore, not only is the coordinate calculation simple, but especially when the target is a binary image, the processing can be further speeded up by using a device that can read and write N vertically or horizontally continuous pixels as one word. An example of the configuration of an apparatus that satisfies this condition will be described below.

FIG. 3 shows the overall configuration of a device that implements the present invention. In the figure, 101 is a processing unit consisting of a microprocessor or a microprogram controlled processor, and 102 is a processing unit 1.
01 is a program memory for storing programs and numerical data such as tables, 103 is an image input device, 104 is an image output device, 105 is an image memory for storing image data, and 106 is a bus for data, addresses, and control signals.

Image memory 105 has the following functions.

(B-1) With the specified pixel coordinates as the left edge or top edge,
N pixels consecutive in the specified access direction (horizontal or vertical) are accessed simultaneously. Here, N indicates the length of data (word length) on the data bus.

The correspondence between pits and pixels within one word is as follows at both ends:
l, 313 (I, east 31gn1f 1ca
The Iri order corresponds to ntdanit).

(B-2) The specified pixel coordinates in the above (H-1) are specified as horizontal and vertical two-dimensional addresses.

The following description will be made assuming that the word length N is 4 for the sake of simplicity.

In order to realize the above (B-1), the image memory is composed of four memory modules, so that four consecutive pixels in any horizontal or vertical direction are stored in different memory modules, respectively. Associate pixels with memory modules. When defining the upper left of the image as the origin, the right horizontal direction as the X axis, and the lower vertical direction as the y axis, the memory module number μ that should store the pixel with coordinates (x, y) is μ = (x + y) // 4 (The correspondence between the pixel and the memory module can be satisfied if If α is α−mXy+(X/4)
The correspondence between pixels and addresses within the memory module can be determined. Here, [ ] is converted into an integer by rounding down the decimal places, and m is a constant specific to the image memory where 4×m is the width of the image memory. FIG. 4(a) schematically shows the correspondence between pixels and memory module numbers according to formula Qa), and FIG. 4(b) schematically shows the correspondence between pixels and addresses in the memory module according to formula (11). In the figure, 201 is a pixel with X coordinates 2 and 3, and is stored at address 6 of the first memory module. however,
Let m be 2.

FIG. 5 shows the configuration of the image memory. In the figure, 301-30
3 is specified in function (B-1). Registers 304 to 307 each store the pixel coordinate X coordinate, the X coordinate, and the access direction, respectively. 1st. Second. The third memory modules 308 to 311 are respectively 0.
1st. Second. An address generation unit 312 that generates an address in the third memory module converts read data into a function (B
- a routing section arranged in the pit order described in 1);
Reference numeral 313 is a routing unit that arranges the pit order described in function (B-1) in an order in which the pits are distributed to predetermined memory modules.

The data reading and writing operations of the image memory 105 are as follows. First, determine the X coordinate (hereinafter abbreviated as xo), the X coordinate (hereinafter abbreviated as 'Io), and the access direction (right, left, bottom, or top direction, hereinafter abbreviated as t) of the left or top pixel of the set of pixels to be accessed. , bus 106 and 320, 321° 322 to registers 301, 302° 303, respectively. Address generator 3
08 to 311, when t=0 (access direction is right),
Coordinates (Xor Yo) + (Xo +1. Yo
)+ (Xo +2+ YO)+ (xo +3+ Y
o), when t-1 (access direction is left),
Coordinates (xo+)'O)+(xo 1+YO)+(x
o2+'IQ)+(X03+)'o), when t-2 (access direction is down), the coordinate (Xo
,yo)(Xo,yo+1)+(xo*Yo
+2 )+ (xo + >'o +3),
When t=3 (access direction is up), the coordinates (xOr Y
O)+ (xO+ Yo 1), (Xo lYo 2)
+ For (xo * Yo 3), an address within each memory module is generated according to equation (1o) and equation (11). When reading data, a read command is then issued to each memory module 304 to 307 using a signal 323. Predetermined data is read onto buses 324-327 by signal 323. The routing unit 312 connects buses 324 to 32
The data of 7 is arranged in the order shown in the correspondence between pixels and bits based on Xo + Yo, and is sent to bus 328.
The signal is output to bus 106 through the bus 106. On the other hand, when writing data, the data path 106 sends the data to be written to the routing section (313) via the bus 328.
Next, when a 1-include command is issued through the signal 323, the routing unit 313 arranges the data in the order shown in the correspondence between pixels and bits based on XO+MO, and
~333.

Finally, the write command signal 323 transfers data to the memory modules 304 to 307 via buses 329 to 3, respectively.
I pledge from 33.

Next, a procedure for realizing the rotation of the present invention using the above-described apparatus will be explained. The processing unit 101 executes a program stored in advance in a program memory. FIG. 6 is a schematic flowchart of this program. First, in an image input step, an image is input using the image input device 103. However, it is also possible to target an image that has been subjected to other processing or an image that has been generated by a program of the processing unit 101, and in this case, this step is omitted. Next, transforms t6)-U) are sequentially executed to rotate the image. Finally, in an image output step, the image is output from the image output device 104. However, it is also possible to perform other processing on the rotated image, and in this case, this step is omitted.

Figures 7 to 10 are detailed flowcharts of the 6 steps of conversion.As each flow has a similar format, they will not be explained individually.
A complete explanation will be given below. First, the conversion to be processed and the pixel position of the converted image are initialized to the upper left pixel. Next, the access direction of the image memory, that is, the value of t, is set. Next, from the current converted image pixel position, move N pixels (N bits - 1 word) in the current access direction to the current converted image pixel position,
Forward. Next, the next N on the same access direction line
Transform and move the transformed image pixel location to the pixel location. After repeating the above for one line, the conversion and conversion target image pixel positions are moved to the beginning of the next line. more than 1
Ends the iterative process for each image.

According to this embodiment, in order to perform rotation processing on a binary image, 4
Although it requires several conversions, it can be processed in units of N pixels (N is word length). However, the processing of N pixels is
All that is required is to increment only one of the y coordinates on the transformed image, and to actually transfer the data.

On the other hand, in the conventional affine transformation method, the transformation is performed twice, but it is necessary to increment both the X and y coordinates of the image to be transformed, pixel by pixel. Therefore, the method according to the present invention is faster than the conventional method using affine transformation when N is 4 or more;
By increasing the value, the speed can be further increased. Regarding the image quality of the rotated image, the method of the present invention is equivalent to the method using affine transformation, and is therefore superior to the conventional method using oblique axis transformation.

Although an embodiment using hardware has been described above, the idea of the present invention can be fully realized by computer system software or the like, and can be implemented in an information processing system that involves image and graphic processing.

〔Effect of the invention〕

As described above, according to the present invention, since rotational transformation can be performed by simply updating coordinate data, it is possible to obtain a rotated image of the same quality as the conventional method using affine transformation at a higher speed than before.

Furthermore, it is possible to implement it in a wide range of forms, from dedicated hardware devices to software, and is expected to be useful in a wide variety of fields.

[Brief explanation of drawings]

Fig. 1 is an explanatory diagram showing the principle of the present invention, Fig. 2 is an explanatory diagram explaining the interpolation rules, Fig. 3 is an overall configuration diagram of an embodiment of a device realizing the present invention, and Fig. 4 is an explanatory diagram showing the correspondence between image memory and images in one embodiment of the present invention, FIG. 5 is a configuration diagram of the image memory, FIG. 6 is a schematic flowchart of the rotation program in the embodiment of the present invention, and FIGS. FIG. 10 is a detailed flowchart of each step of the rotation program. 101... Processing unit, 105... Image memory, 304
~307...Memory module, 308~311...
・Address generation unit, 312, 313...Routing unit,...Vertical oblique axis conversion,...Vertical reduction conversion, Figure 1, Figure 2, (a-) χ Figure 6, Figure 7

Claims (1)

[Claims] 1. Obtaining a second image by performing oblique axis transformation from the first image in the first direction; and performing scaling transformation from the second image in the first direction. obtaining a third image;
a step of performing scaling conversion in a second direction from the image to obtain a fourth image; and a step of performing oblique axis conversion in the second direction from the fourth image to obtain a fifth image. An image rotation method characterized by: 2. Consisting of a means for inputting an image, a means for specifying a reference pixel address and a pixel access direction, a means for reading and writing continuous plural pixel data from the pixel in the access direction, and a means for accumulating these pixel data. Image rotation device.