JPS60258688A - Pattern recognizing method and device using the method - Google Patents

Abstract

PURPOSE:To extract feature points without any skeletonizing process and to perform stable pattern recognition by providing each picture element of an input pattern with points corresponding to the number of pattern places which are present arount it, and generating a point-given pattern. CONSTITUTION:A video signal generated by an image pickup device 1 is converted by a binary coding circuit 2 into a binary signal. Then, the binary-coded image data (input image pattern) is stored in a frame memory 3. The image data in the memory 3 is used by a maximum line width measuring circuit 4 to calculate maximum line width (maximum character width). The calculated maximum character width is supplied to a window outer peripheral character place counting circuit 5 to give points to each character bit, and the result is outputted to a point-given pattern memory 6. For the purpose, character bits which do not connect with corresponding to character bits in a specific window are erased. Then, the number of character areas (pattern plates) of the outer periphery part of the window is counted.

Description

Detailed Description of the Invention (a) Technical Field of the Invention The present invention relates to a pattern recognition method and a device using the same, and particularly relates to a pattern recognition method and a device using the same. The present invention relates to a pattern recognition method for recognizing an input image pattern by converting it into a digitized image, and a device using the same.

(0) Background of the technology Improving the recognition rate of pattern recognition devices is an important issue, and for this purpose, it is necessary to be able to successfully extract feature points.

Conventional methods for extracting feature points have the drawback that there is a high possibility that undesirable elements may be mixed in with the extraction, which is an obstacle to pattern recognition.Therefore, a technology that can avoid such problems is needed. Development of a method is required.

(C) Prior Art and Problems In conventional character recognition, character feature points are extracted after character skeletonization processing. In the skeletonization process, as shown in (1-1) in Figure 1, whiskers a appear in areas other than the core line of the character, and as shown in (1-2) in Figure 1, the intersection of two straight lines becomes two triangles. There is a problem that stable character recognition is hindered because the characters are skeletonized as branching points.

(d) Purpose of the Invention The present invention was devised in view of the drawbacks of the conventional techniques as described above. An object of the present invention is to provide a pattern recognition method capable of performing pattern recognition using the same method, and an apparatus using the same.

(B) Structure of the Invention In order to achieve this objective, the method according to the present invention assigns a score corresponding to the number of pattern parts existing around each pixel of an input image pattern obtained by imaging. A first device of the present invention generates a scored pattern and uses the scored pattern to recognize the input image pattern, and a first device of the present invention includes an imaging device, a binarization circuit connected to the imaging device, and a binarization circuit connected to the imaging device. a frame memory connected to the binarization circuit; a maximum line width measuring circuit connected to the frame memory and measuring the maximum line width of the image pattern; and a maximum line width measuring circuit connected to the frame memory and the maximum line width measuring circuit, A scored pattern generation circuit that positions the center of the window determined by the width at the pixel to be scored, counts the number of pattern locations on the outer periphery of the window, and generates a scored pattern with the counted number of points as the score of the pixel. and a scored pattern memory connected to the scored pattern generation circuit, a dictionary memory, and a recognition logic device connected to the scored pattern memory and the dictionary memory, The device includes an imaging device, a binarization circuit connected to the imaging device, a frame memory connected to the binarization circuit, and a maximum line width connected to the frame memory that calculates the maximum line width of the image pattern. The measurement circuit is connected to the frame memory and the maximum line width measurement circuit, and the center of the window determined by the maximum line width is positioned at the pixel to be converted into points, and the number of pattern locations on the outer periphery of the window is counted. A scored pattern generation circuit that generates a scored pattern with the score as the number of the pixel, a scored pattern memory connected to the scored pattern generation circuit, and a scored pattern memory connected to the scored pattern memory that generates the center of gravity of the feature region. The center of gravity determining circuit includes a center of gravity determining circuit for determining the center of gravity, a dictionary memory, and a recognition logic device connected to the center of gravity determining circuit and the dictionary memory.

(→Examples of the Invention The present invention will be described in detail below with reference to the accompanying drawings.

FIG. 2 shows one embodiment of this sale. The pattern recognition device of this embodiment includes an imaging device 1, a binarization circuit 2, a frame memory 3, a maximum line width measurement circuit 4, a window outer periphery character portion counting circuit 5, a pattern memory with points 6, a dictionary memory 7, and an epistemology. Consists of embedded W8. Next, the operation of the apparatus of the present invention having this configuration will be explained.

The video signal captured by the imaging device 1 is binarized by the binarization circuit 2, and the binarized image data (input image Bakun) is
is stored in the frame memory 3.

Then, the maximum line width is determined by using the image data in the memory 3 in the maximum line width measuring circuit 4 (Stenbu Sl in FIG. 3). This will be explained with reference to FIG. First, a 2 x 2 window is scanned over the image data, and when a point is detected where the window is completely included in the text area, scanning with that window is finished and the window is changed to 3 x 3. and perform a similar scan. In this way, the size of the window is increased by 1 bit (pixel), and when there is no place where the window can be completely included in the character area, the size of the window is (n + 1) x (n + 1). If there is, the maximum character width of that character part is defined as n. Regarding Figure 4, the 3 x 3 window is completely included in the text area, but the 4 x 4 window is not completely included in the text area, so the maximum line width in this text area is (maximum character width)
is 3.

The maximum character width determined in this way is used in the window outer periphery character place counting circuit 5 to score all character bits in the following manner, and the result is output to the scored pattern memory 6.

First, in order to score the character bits and 7 bits (maximum character width)
Position the center of the 2+1 window at the character bit to be scored, and erase the character focus that is not continuous with the character bit within the window (see step S2 in Figure 3 and Figure 5). . After that, the number of character areas (pattern locations) on the outer periphery of the window is counted. As shown in Figure 6, the counting method is to scan the outer circumference of the positioned window clockwise and add 1 to the counter when moving from the background area to the character area, or to add 1 to the counter when moving from the background area to the character area. When moving to the background part, 1 may be added to the counter, or both of these may be counted and divided by 2 after the second cycle.

The count thus obtained is used as the score of the character bit (character focus of interest) to be scored (step S3 in FIG. 3).

When points are assigned in this way, each character bit is assigned a score of 1 for an end point, 2 for a point on the line, 3 for a three-way junction, etc., as shown in Figure 7. It is possible to define each region using the following methods, and extract characteristic regions.

When such a scoring process is performed, for an n-junction point that has four or more branches, there is always an area of (n-1> points) around the n-point area, as shown in Figure 8. This (n-1) point area is considered to be a part of the n point area and is replaced with the n point (Fig. 8, step S4 in Fig. 3).
).

When the feature region extraction is completed as described above (step S5 in FIG. 3), the resulting scored pattern is output to the scored pattern memory 6. An example of the scoring pattern is shown in FIG.

This scored pattern is supplied to the recognition logic device 8 together with the dictionary scored pattern in the dictionary memory 7 for pattern recognition.

FIG. 10 shows another embodiment of the present invention, in which the structure up to the pattern memory 6 with scores is the same as that of the embodiment shown in FIG.
A center of gravity determining circuit 9 is provided between the circuit 9 and the output of the circuit 9.
This differs from the embodiment shown in FIG. 2 in that the output of the stored dictionary memory 7'' is connected to a recognition logic device 8'.

By providing this center of gravity determination circuit 9, the center of gravity of each feature region can be determined, and feature points can be extracted with the feature regions as points.

In the above embodiment, an example in which the window is square has been described, but as is clear from the above description, the window is not limited to this.

(G) Effects of the Invention As described above, according to the present invention, the above-mentioned scoring process is performed on the input image pattern used for pattern recognition, so it is not possible to use the conventional skeletonization process. Problems that were difficult to avoid can be eliminated, and patterns can be recognized stably.

[Brief explanation of the drawing]

Fig. 1 is a diagram for explaining problems that occur in conventional skeletonization processing, Fig. 2 is a diagram showing one embodiment of the present invention, and Fig. 3 is a diagram showing the main part of the embodiment shown in Fig. 2. Flowchart, Figure 4 is a diagram used to explain the maximum line width measurement of a pattern,
Figure 5 is a diagram illustrating the erasure of character bits that are not contiguous with the focused point for scoring, Figure 6 is a diagram used to explain the method of counting points, and Figure 7 is for windows FIG. 8 is a diagram illustrating the manner of the pattern occupied by a branch point and the allocation of points, FIG. , FIG. 10 is a diagram illustrating another embodiment of the present invention. In the figure, 1 is an imaging device, 2 is a binarization circuit, 3 is a frame memory, 4 is a maximum line width measurement circuit, 5 is a window outer circumference character position counting circuit, 6 is a Bakun memory with points, 7 is a dictionary memory, 8
is an epistemic logic device. Figure 1 (1-1) (1-2) Nii Nao, Wataru L44 No. 辷訃梧It Nii Elephant *+ This book is missing Figure 3 Figure 4 Figure 5 Figure 6 Figure 7 11M Swallow 1. Gu, Beach of system tL1, 2 points Figure 8 Figure 10 Figure 9 Window size' 5

Claims (1)

[Claims] +1) A scored pattern is generated by assigning points to each pixel of an input image pattern obtained by imaging in accordance with the number of Bakun points existing around it, and the scored pattern is A pattern recognition method in which the above is used to recognize the input image pattern. (2) An imaging device, a binarization circuit connected to the imaging device, a frame memory connected to the binarization circuit, and a maximum line connected to the frame memory for measuring the maximum line width of the image pattern. The width measuring circuit is connected to the frame memory and the maximum line width measuring circuit, and the center of the window determined by the maximum line width is positioned at the pixel to be scored, and the number of pattern locations on the outer periphery of the window is counted. a scored pattern generation circuit that generates a scored pattern with the counted number of points as the score of the pixel, a scored pattern memory connected to the scored pattern generation circuit, a dictionary memory, the scored pattern memory, and 1. A pattern recognition device comprising: a recognition logic device connected to a dictionary memory. (3) An imaging device, a binarization circuit connected to the imaging device, a frame memory connected to the binarization circuit, and a maximum line connected to the frame memory for measuring the maximum line width of the image pattern. The width measuring circuit is connected to the frame memory and the maximum line width measuring circuit, and the center of the window determined by the maximum line width is positioned at the pixel to be converted into points, and the number of pattern locations on the outer periphery of the window is counted. A scored pattern generation circuit that generates a scored pattern with the counted number of points as the score of the pixel, a scored pattern memory connected to the scored pattern generation circuit, and a scored pattern memory connected to the scored pattern memory that A pattern recognition device comprising: a center of gravity determining circuit for determining the center of gravity; a dictionary memory; and a recognition logic device connected to the center of gravity determining circuit and the dictionary memory.