JPH0756673B2 - Image processing method by divided space filter - Google Patents

Description

The present invention relates to a space filter in an image processing device,
The present invention relates to an image processing method using a divided space filter suitable for executing a space filter having an arbitrary shape and arbitrary size.

[Conventional technology]

As described in Japanese Patent Laid-Open No. 146366/1984, a conventional device is composed of a delay circuit by a raster scan and a line buffer, a spatial product sum operation circuit, etc. (FIG. 2). This formula can process at a very high speed when constructing an n × m space filter, but has a drawback that the amount of hardware increases in proportion to the area of the space filter.

On the other hand, even with this formula, it is possible to realize an n × n spatial filter by repeating the processing of a small area spatial filter (for example, 3 × 3) without increasing the amount of hardware, but the processing time is a spatial filter. There was a drawback that it increased in proportion to the area.

In FIG. 2, 6 is a processing target image, 7 is a processing result image, 8 is a raster scan raster, 9 is an example of a 3 × 3 spatial filter, 10 is a 3 × 3 spatial filter target pixel, and 11 is a spatial filter target pixel. 3x
The three spatial filter processing result pixels and the spatial product sum operation unit 12 are shown.

[Problems to be solved by the invention]

In the prior art, the spatial filter used for image processing is n × n.
For example, if the processing speed is prioritized, the amount of hardware will increase in proportion to the area of the space filter, and if the reduction of the amount of hardware is prioritized, the processing speed will increase. There is a drawback that it increases in proportion to the area.

An object of the present invention is to provide a divided space filter system which gives priority to reduction of the amount of hardware and can improve the processing speed.

[Means for solving problems]

The purpose is to arrange a space filter having an area of, for example, 3 × 3 in an arbitrary space as shown in FIG. 1 to construct a space filter of an arbitrary shape, and as shown in FIG. It can be achieved by executing a small area space filter and shifting and adding the output result to the target pixels of the whole space filter of arbitrary shape. In FIG. 1, 1 is an example of an n × n space filter. 2 is the target pixel of the n × m space filter. 3a to 3h are 3 × 3 space filters. Reference numeral 4 denotes a target pixel of the 3 × 3 spatial filter, and 5a to 5h denote shift amounts of the 3 × 3 spatial filter processed image.

In this case, place the space filters in the small area in orderly
If arranged as shown in the figure, it is also possible to construct an n × m large area space filter.

That is, when 14 is an n × m space filter, 15a and 15b represent 3 × 3 space filter processing target pixels, and 16 represents an n × m space filter processing target pixel.

If there are many “0” data in the data structure of this n × m large area space filter (FIG. 4), the small area space filter arranged there does not need to be calculated, so that FIG. As shown in (3), a space filter (corresponding to a hatching portion) is obtained by thinning out the "0" data portion, and the calculation speed is improved.

In FIG. 4, 17 is a part other than zero data, 18 is a 0 data part, and 19 is an n × m spatial filter.

Further, in FIG. 5, 20 is an n × m space filter, and 21 is 3 × 3.
Represents the space filter of.

Further, by arranging the small area space filters spatially so as to overlap each other, it is possible to realize a synergistic effect that cannot be realized by a normal n × m space filter.

[Action]

Generally, a spatial filter has pixels for spatial filter. For example, in a 3 × 3 spatial filter, the center is usually set to the spatial filter target pixel. However, there is no problem mathematically or physically even if the target pixel is not the center. This is equivalent to constructing an n × m space filter by the arrangement of the target pixel and the applied space filter, as shown in FIG.

In FIG. 6, 22 is an n × m spatial filter, 23 is a 3 × 3 spatial filter, 24 is an n × m spatial filter processing target pixel, and 25 is a 0 data part. However, the data other than the applied spatial filter, on the n × m spatial filter,
There is a condition that they are equivalent to all being "0".

If this idea is expanded, a spatial filter of an arbitrary shape can be easily constructed by the theory of superposition as shown in FIG. 26 is an n × m space filter, 27a, 27 as described above.
Reference numerals b and 27c represent 3 × 3 space filters, and 28 represents pixels subject to n × m space filter processing. And 17a-c can be piled up and a filter of arbitrary shape can be comprised.

By the way, selecting a target pixel other than the center of each spatial filter means executing a spatial filter in which the target pixel is set at the center of the spatial filter as shown in FIG. 8 and outputting the output result to an arbitrary target pixel. It is equivalent in the effective area to shift by the deviation of.

Based on this idea, the construction method of the space filter of arbitrary shape is
Here, it is called a divided space filter method.

As described above, this division space filter system can be combined to construct a space filter for a large area by combining small area space filters of about 3 × 3 with mosaics or tiles attached to necessary areas, and Can be realized without increasing.

In FIG. 8, 29 is a processing target pixel of the 3 × 3 spatial filter, 30 is a 3 × 3 spatial filter, 31 is an n × m spatial filtering processing pixel, 32 is an n × m spatial filter, and 33 is n × n. m
, 34 is an effective data range after execution of the spatial filter, 34 is an original image data area, 35 is an effective data range after execution of the 3 × 3 spatial filter, and 36 is a shift data area of 35 images.

〔Example〕

An embodiment of the present invention will be described below with reference to FIGS. 9 to 15.

The image to be processed by the image processing method of the present invention is an image scanned by the raster scan method.

(See FIG. 2) FIG. 9 shows the hardware configuration. An ITV camera 37 for inputting image data, a monitor TV 38 for outputting (displaying) image data, an image memory 39 for storing image data,
Image processing processor 40 for executing 3 × 3 space filter
(For details, refer to JP-A-60-53349 "Image processing processor" and FIG. 2) and a console CRT of Man-Machine Interface. Reference numeral 41 is an image processing device.

FIGS. 10A and 10B show a combination of an n × m large area space filter and an equivalent 3 × 3 space filter executed. 42 indicates an n × m space filter, 43 indicates a 0 data portion, and 44 indicates a data other than 0 data.

45 is the pixel for n × m space filter processing, 46 and 47 are 3 × 3
The spatial filters (1) and (2) are shown, 48 is a combination of 3 × 3 spatial filters, and 49 is a 3 × 3 spatial filter (n).

The execution procedure for this is shown in FIG. First, 3 shown in FIG.
The × 3 spatial filter number (1) (that is, 50 (1)) is executed, and the resulting image is shifted by the deviation between the target pixel position of the large area spatial filter and the target pixel position of the 3 × 3 spatial filter. Store in memory. Then, the spatial filter number (2) (that is, 50 (2)) is executed, the resulting image is similarly shifted, and the shifted image and the shifted result image of the spatial filter number (1) are added. Thereafter, the process is repeated in the same manner until the space filter number (k). At this time, the finally obtained processing result image is
The result is the same as that of applying the n × m large area space filter shown in the tenth.

The reason for this is that the spatial filter arithmetic processing is a linear arithmetic operation (see the equation (1)). However, k ₁ + k ₂ + 1 = n l ₁ + l ₂ + 1 = mu is 0 to (n / 3-3), v is 0 to (n / 3-3) S _p , _q are p and y in the x direction. It is defined as an operator that shifts by q in the direction.

Is a 3 × 3 spatial filter arithmetic expression, f _{x + 1 + i + j} is a pixel to be processed, w _{x + 1} and _{i + j} are weighting factors of the spatial filter, and g ′ _x and _y are 3 ×. 3, spatial filter operation results, g _x , _y indicate n × m spatial filter operation results, and x, y, x + i, x + j, etc. indicate coordinates.

Equation (1) holds when n, m of n × m is a multiple of 3, but when it is not a multiple of 3, the remaining aria is 2 ×
Fill (replenish) with a combination of 2,2 × 1,1 × 2,1 × 1. (See formula (2)) However, k ₁ , k ₂ , l ₁ , l ₂ , u, v, S _p , _q are the same as in formula (1). As can be seen from the above explanation, Fig. 11 shows 50 ( ₁ ), 52
( ₂ ) indicates 3 × 3 space filters (1) and (2) respectively, 51 is the original image data area, 53 is the effective area after the 3 × 3 space filter, and 54 ( ₁ ), 54 ( ₂ ) are Indicates the deviation shift amount. Reference numeral 55 indicates addition processing, and 56 indicates the effective area after addition.

On the other hand, when trying to realize n × m by hardware, the number of image processors increases by (n / 3 × m / 3) times as shown in FIG. (Refer to FIG. 2 for an image processing processor that realizes a 3 × 3 spatial filter.) In the present invention, one image processing processor is used for n × m image processing without using a plurality of image processing processor groups as shown in FIG. Since a space filter can be realized, the hardware can be greatly reduced.

In FIG. 12, 66 is an input data line, 67 is an image processing processor, 68 is an output data line, 69 is a calculation result data line, 70 is an adder, and 71 is an addition result data line.

Further, if a function of shifting the result image data of the 3 × 3 spatial filter and a function of adding the shift result image of another 3 × 3 spatial filter in parallel can be provided, the speed can be further increased. Specify the concrete hardware configuration
Shown in Figure 13.

72 is an image processing area, 73 is a processing target pixel, 74 is a raster scan, 75 is a spatial filter, 76 is an image processing target image, 77
Is an image processing result image, 78 is an image processing processor, 79 is a shift calculator, and 80 is an adder.

Next, the shortening of the processing speed, which is the greatest feature of the present invention, will be described.

As can be seen from FIG. 12, if an n × m space filter is to be executed, the calculation of the 3 × 3 space filter is (n / 3 × m
/ 3) Must be repeated. In other words, if this is executed by one image processing processor, it takes (n / 3 × m / 3) times as long. However, if the divided spatial filter, which is the core of the present invention, is used, only a meaningful spatial filter (a spatial filter that has no 0 data) needs to be calculated. Therefore, the image processing time can be reduced by the ratio shown in Expression (3). It is possible to

In other words, {area of space filter of arbitrary shape (3 × 3
The speed can be increased by the ratio of (the multiple of the space filter of)): {area of the space filter of n × m}. (No. 1,4,5,
(Refer to 10.) Furthermore, if the characteristics of the spatial filter are used, a higher speed can be achieved. For example, taking the space filter of FIG. 14 (a) as an example, this 9 × 9 space filter can be realized by using nine space filters in the divided space filter of the present invention, but further speedup is intended. You can also That is, it can be realized by using one filter shown in FIG. 14 (b).

This is clear from equations (4) and (5).

However, it should be noted that when adding, it is necessary to multiply by 8 (in the case of formula (4)) or (-1) times (in the case of formula (5)). Note that (-1) times can be further speeded up by changing (multiplication + addition) to only subtraction.

By using the equations (4) and (5), the spatial filter is calculated once, and after that, the 9 × 9 spatial filter of FIG. 14 (a) can be obtained by only shifting and adding (or subtracting). It can be realized. The time reduction ratio at this time is 1/9.

Finally, the synergistic effect which is another feature of the present invention will be described. This is particularly effective for pattern matching of grayscale images.

For example, in the case of a quadrangular object as shown in FIG. 15 (a), the spatial filters of the area are overlapped and arranged in the vicinity of the vertices which are considered to have the most effective features as a graphic, and the target pixel is quadrangular. By setting it at the center, the position of the object can be detected. (FIG. 15 (b)) In a uniform n × m space filter, both a surface with few features (for example, a surface illustrated by a quadrangle) and a feature point (for example, a vertex) are processed at the same level. Therefore, the information on the characteristic points is diffused (thinned).

The degree of diffusion is a ratio of (3 × 3 area) :( n × m area).

In order to eliminate this drawback, by placing the spatial filters with emphasis on the most characteristic points such as vertices, the spatial filters that are not overlapped work more strongly (in the sense of extracting the features). It will be. This is called the synergistic effect here.

The small area space filter in the synergistic effect example is a differential system (X
Directional differentiation, Y-direction differentiation, etc.) are particularly effective.

In FIGS. 15A and 15B, reference numeral 91 is a processing target image, 92 is a quadrangle (target object), 93 is a small area space filter, and 94 is a target pixel.

〔The invention's effect〕

According to the present invention, a spatial filter of arbitrary shape, for example 3
By combining (repeating) small area space filters of × 3 (repetition processing), compared to the processing time of n × m space filters (area of space filter of arbitrary shape): (area of n × m space filter including this) There is an effect of shortening at a rate of.

[Brief description of drawings]

FIG. 1 is an outline of a divided space filter system, FIG. 2 is an outline of a conventional space filter system according to the prior art, FIGS. 3 to 8 are principle explanatory diagrams of the divided space filter system, and FIGS. It is a figure. Fig. 12 is an example of implementation with hardware, Fig. 13 is an explanatory diagram of the shift function,
FIGS. 14 (a) and 14 (b) show examples of space filters, and FIGS. 15 (a) and 15 (b) show examples of actual objects and space filters. 1 ... n × m space filter, 2 ... target pixel, 3a to 3h ...
… 3 × 3 space filter, 5a-5h …… Shift amount of 3 × 3 space filter.

Claims (1)

[Claims] 1. A grayscale image data processing method in an image processing apparatus comprising a grayscale image data input means for inputting a grayscale image and a grayscale image processing means, wherein m × n pixels of vertical pixel n and horizontal pixel m. A plurality of unit spatial filters partially consisting of unit pixels smaller than m × n pixels are set in the space according to the characteristics of the grayscale image input from the grayscale image data input means, and the grayscale image data input means is set. The plurality of unit spatial filters are respectively acted on the grayscale image input by, and the operation results of the plurality of unit spatial filters are gathered at a specific target pixel in the m × n pixel space, An image processing method using a divided spatial filter, wherein the action results of the plurality of unit spatial filters are shifted and added.