DE68924910T2 - Image processing. - Google Patents

Description

When recognizing the handwritten letter "R" as shown in Fig. 1, the concaves C1 and C2, and the opening H are the key points in recognition. Concaves, convexes, and openings represent significant characteristics of a configuration. By extracting these characteristics, for example, a convex envelope is placed over the configuration and the original configuration is subtracted from this convex envelope. However, in a digital image, the convex envelope cannot always be generated precisely. In a digital image, the convex envelope is generated by successively connecting extreme points with lines and by filling the inside of the lines. However, the line differs in a digital image according to the algorithm used for line generation. This means that there may be a deviation of one pixel because an analog line does not necessarily pass through the center of a digital pixel. If a pixel is selected that differs significantly from the analog convex envelope, the configuration produced by subtracting the original image does not exactly represent the characteristics of the character. Typically, a digital line is drawn by a graphics processor, i.e. according to a processor's algorithm.

If the convex envelope is not generated accurately, the analysis after the mentioned subtraction cannot be carried out properly.

In PROCEEDINGS OF THE FALL JOINT COMPUTER CONFERENCE, San Francisco, California, 9-11 December 1968 (AFIPS Conf. Proceedings, Vol. 33) pages 1125 to 1138; JH MUNSON: "Experiment in the recognition of hand-printed text: Part 1 - Character recognition", it was proposed that in order to produce accurate recognition of a hand-printed character, a convex hull boundary (CHB) should be placed around the character, this being achieved by determining extremal points of the character image and connecting these extremal points by straight lines, a straight line between two extremal points following as closely as possible a theoretical straight line between the two extremal points, but never falling outside the theoretical straight line. After the CHB is obtained, concave spots and inclusions in the character image are found using suitable computer routines.

In COMP. GRAPH. & IMAGE PROC., Vol. 9, No. 2, February 1979, Pages 183 to 195, Academic Press Inc.; R. L. T. CEDERBERG: "A new method for vector generation" an algorithm for calculating constants in an equation that defines the smoothest digital representation of a straight line is proposed with respect to the Bresenham algorithm. In particular, it is proposed to approximate a theoretical straight line by first determining a basic chain code group that approximates a part of the straight line and then repeating the found basic chain code group according to the length of the straight line to be approximated.

It is the object of the present invention to provide an improved method for recognizing a configuration, by which it is possible to generate a representation of a convex envelope boundary of the configuration and whereby between two extremal pixels the theoretical straight line connecting the extremal pixels is approximated as closely as possible, but never the area outside the theoretical straight line is touched.

This problem is solved by the subject matter of patent claim 1.

A preferred embodiment of the present invention is explained in more detail below with reference to the accompanying drawings, in which the drawings show in detail:

Fig. 1 a handwritten letter "R";

Fig. 2 shows the definition of a chain code;

Fig. 3 a configuration with a concave point; and

Fig. 4 shows an image divided into four quadrants.

As shown in Fig. 2, a pixel in a digital image has eight peripheral pixels, so that there are eight directions from the pixel leading to adjacent pixels. These directions are described by a Freeman chain code from "0" to "7". A sequence of chain codes represents a continuous pixel train, as well as the deviation of each pixel from a line.

Fig. 3 shows a concave point that has extremal points a and b. The points a and b lie on the convex envelope, and between these points lie boundary pixels C1 to C9. A reference line 1 represents the analog convex envelope. The digital convex envelope should never fall outside the theoretical line 1 and should follow the theoretical line 1 as closely as possible. The corresponding pixels are determined one after the other starting from an extremal point, for example point a. The pixel order can be clockwise or counterclockwise. For the following description, the counterclockwise direction is assumed.

The chain code assigned to the extremal point a is "4" or "5" and cannot be any other code. The pixel d', indicated by "4", is closer to line 1 than the pixel d1, indicated by the chain code "5". The pixel d'is but outside line 1. Therefore, pixel d1 is selected as the pixel adjacent to extremal point a for generating the convex envelope. In the same way, pixels from d2 to d8 are selected one after another until extremal point b is reached. Line 1 defines the coordinates of extremal points a and b. The distance between line 1 and each pixel between d1 and d8 is defined by the chain code train as specified by pixels d1 to d8. The inclination angle θ of the line is specified by the coordinates of extremal points a and b. The distance between line 1 and each pixel between d1 and d8 is specified by the inclination angle θ and the chain code train. The distance Δ between line 1 and pixel d is calculated as follows, given a chain code of "5":

Δ = (δ/sin 45º)sin(45º - θ)

= -δ(sinθ - cosθ),

where

δ indicates the width of a pixel, as well as the distance between vertically and horizontally spaced pixels;

Δ becomes negative if the pixel is outside line 1. The value Δ for the pixels after pixel d1 results in a distance increase or decrease. If the value Δ for d1 is Δd1 and for d2 is Δd2, the distance between the line and pixel d2 will be (Δd1 + Δd2). The value Δ for each chain code can be calculated in advance. The pixels are selected so that the distance becomes minimal and never becomes negative.

A configuration can be divided into four quadrants. If the upper end pixels of the configuration in Fig. 4 are designated P0 and P7, the lower end pixels P3 and P4, the left pixels P1 and P2, and the right pixels P5 and P6, in the areas between P0 and P2, P2 and P4, P4 and P6, and P6 and P0, the pitch increase or decrease can be determined only by the chain code. The area from P0 to P2, P2 to P4, P4 to P6, and P6 to P0 are hereinafter called the first, second, third, and fourth quadrants. The pitch increase or decrease in each quadrant is shown in Table 1. In Table 1, the slope θ of the line is measured counterclockwise from a horizontal line, and δ indicates the distance between horizontal and vertical adjacent pixels. Table 1 Chain Code Quadrant

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

d1: -δ(sinθ-cosθ)

d2: -δ(sinθ-cosθ)-δsinθ

d3: -δ(sinθ-cosθ)-2δsinθ

d4: -δ(sinθ-cosθ)-3δsinθ

d5: -δ(sinθ-cosθ)-4δsinθ

d6: -δ(sinθ-cosθ)-5δsinθ

d7: -δ(sinθ-cosθ)-6δsinθ

d8: -δ(sinθ-cosθ)-7δsinθ

The distance increase or decrease in Table 1 can be easily calculated in advance if θ is determined. Thus, no trigonometric calculation is necessary for each pixel.

Claims (3)

1. A method for recognizing a configuration stored pixel by pixel in an image processing device, the method comprising generating chain code train segments for describing a CHB of the configuration and comprising the generation of the following steps: Determining extremal pixels of the configuration which lie on the CHB; Calculating straight lines between neighboring extremal pixels; approximating each straight line by a line defined by a chain code train segment, the line following the straight line as closely as possible but never falling outside the area between the straight line and the configuration, characterized in that the selection of each individual chain code element of a chain code train segment is achieved by calculating the distance between the corresponding straight line and the top point of the line as defined by the last chain code element of the elements of the chain code train segment that have already been selected, and by choosing the next chain code element under the criterion of minimizing the distance. 2. The method of claim 1, wherein the distance increment or decrement caused by each different chain code element is calculated before the chain code train section is generated. 3. The method of claim 1 or 2, wherein the calculating includes calculating the inclination angle of the straight line.