Image Processing Method

the present invention relates to an image processing method, particularly to an image processing method for generating a convex hull of a digital image.

In the case of recognizing a hand-written character "R" as shown in FIG. 1, the concavities C1 and C2 and the hole H will be the key points for recognition. It will be understood that the concavities, convexities and holes are important characteristics of a configuration. For extracting these characteristics, for example, a convex hull of a configuration is generated and the original configuration is subtracted from the convex hull. However, in the digital image, the convex hull is not always precisely generated. In the digital image, the convex hull is generated by connecting top points successively with lines and by filling the inside of the lines. However, the line in the digital image differs according to the algorithm of line drawing. Usually one dot deviation may easily happen because an analog line does not always pass through the center of a digital pixel. When a pixel is selected, as a pixel on the digital line, from outside the pixel of the analog convex hull, the configuration generated by the subtraction of the original image from the convex hull does not represent an identifying characteristic itself. Usually a digital line is drawn by a graphic processor, so the digital line is drawn according to the algorithm of the processor.

When the convex hull is not drawn precisely, the analysis after the above subtraction cannot be well performed.

Therefore, the present invention has an object to provide an image processing method for generating a convex hull useful for the analysis of concavities, convexities and holes.

According to the present invention, between two adjacent top points in the convex hull, a pixel inside of and with nearest distance from the line connecting the top points is successively selected, from one top point to the other.

FIG. 1 shows a hand-written character "R";

FIG. 2 is a diagram showing chain code;

FIG. 3 is a diagram showing a concavity of a configuration;

FIG. 4 shows 4 quadrants of an image; and

FIG. 5 shows a flow chart of a method in accordance with the present invention.

Exemplary preferred embodiments of the image processing method according to the present invention will be described in detail hereinafter.

As shown in FIG. 2, a pixel in a digital image has 8 peripheral pixels, so there are 8 directions from one pixel to an adjacent pixel. These directions are represented by Freeman's chain codes, from "0" to "7". A chain code string represents a continuous train of pixels as well as the deviation of each pixel from a line.

FIG. 3 shows a concavity of a configuration consisting of top points a and b on the convex hull and boundary pixels from C1 to C9. A reference line 1 is drawn for representing the analog convex hull (see also item 20 of FIG. 5). When the digital convex hull is generated, pixels should be put inside the line l. The pixels to be selected should also be the nearest to the line l. Such pixels are selected continuously and successively from one top point, for example the point a, to the other (see also items 24 and 30 of FIG. 5). Pixel order may be clockwise or anticlockwise. The anticlockwise order is applied in the following description.

A chain code to be given to the top point a is "4" or "5" and cannot be other codes. The pixel d' indicated by "4" is nearer than the pixel d1 indicated by "5" to the line l. However, the pixel d' is outside of the line l. Therefore, the pixel d1 is selected as a pixel adjacent to the top point a for generating the convex hull. In the same manner, pixels from d2 to d8 are continuously and successively selected toward the top point b. The line l is defined by the coordinates of the top points a and b. The distance between the line l and each pixel of d1 to d8 is defend by the chain code train given to the pixels from d1 to d8. The inclination angle .theta. of the line l is defined by the coordinates of the top points a and b. The distance between the line l and each pixel of d1 to d8 is defined by the inclination angle .theta. and the chain code train (see also items 26 of FIG. 5). The distance .DELTA. between the line l and the pixel d1 is calculated when the chain code is "5" as follows: ##EQU1## where, .delta.: Width of a pixel as well as distance between vertically or horizontally adjacent pixels;

.DELTA. is negative when the pixel is outside of the line l.

The value .DELTA. for the pixels after the pixel d1 gives distance increase or decrease. When the value .DELTA. for d1 is .DELTA.d1 and for d2 is .DELTA.d2, the distance between the line and the pixel d2 is (.DELTA.d1+.DELTA.d2). The value .DELTA. for each chain code may be calculated beforehand (see also item 22 of FIG. 5). The pixels are selected so that the distance is minimum as far as the distance does not become negative, with tracing selected pixels (see also item 28 of FIG. 5).

A configuration can be divided into four quadrants, from the first to fourth. When upper end pixels of the configuration in FIG. 4 are designated P0 and P7, lower end pixels P3 and P4, left end pixels P1 and P2 and right end pixels P5 and P6, in each of the areas, from P0 to P2, from P2 to P4, from P4 to P6, and from P6 to P0, the distance increase or decrease is defined only by the chain code. The area P0 to P2, P2 to P4, P4 to P6 and P6 to P0 are now called the first, second, third and fourth quadrant, respectively. The distance increase or decrease in each quadrant is shown in Table 1. In Table 1, the inclination .theta. of the line is measured anticlockwisely from a horizontal line and .delta. designates the distance between horizontally or vertically adjacent pixels.

In FIG. 3, distances from the line 1 to pixels of d1 to d8 are calculated as follows:

The distance increase or decrease in Table 1 can be easily calculated beforehand when .theta. is determined. Therefore, the trigonometric calculation is not necessary for each pixel.