JPS63196993A - Loop detecting method for optical character reader - Google Patents

Abstract

PURPOSE:To detect the number of loops of a character graphic pattern by a simple algorithm, by detecting the number of loops in a character graphic, based on a confluence of the same stroke. CONSTITUTION:In a pattern memory 1, a pattern 2 which becomes an object to be recognized is stored, and its contents can be referred to arbitrarily by an X coordinate and a Y coordinate shown by an X counter 3 and a Y counter 4. The pattern 2 is brought to a raster scan once extending over the whole surface, and in each main scan, a coordinate by which the pattern is varied from white to black and from black to white is detected. Also, a variation point coordinate obtained in some main scan is compared with a variation point coordinate obtained in the previous main scan, the continuity of a stroke is checked and it is managed by a table. In this case, the number of confluences of the same stroke corresponds to the number of loops, therefore, by calculating the number of confluences, based on the table, the number of loops of the pattern concerned is calculated.

Description

DETAILED DESCRIPTION OF THE INVENTION (Industrial Application Field) The present invention relates to a method for detecting loops, which are one of the important characteristics of character shapes in character recognition, in an optical character reading device (hereinafter abbreviated as OCR). It is something.

(Prior art) When performing character recognition using OCR, first, a character pattern to be read is extracted one character at a time, and the inclination, intersection, curvature, number of character lines, and loops of the character lines are determined for the pattern. Characteristics of the character pattern, such as presence or absence, are extracted, and the combination of these characteristics determines what the character was. Among these characteristics, the number of loops is one important characteristic.

For example, considering the numbers 0 to 9, the number of loops O is 1, 2, 3, 4, 5.7, and the number of loops 1 is 0.
, 6.9, and the number of loops is 8. In other words, this much classification is possible just by the number of loops, and in particular 8 can be recognized just by the number of loops.

The number of loops has conventionally been detected using the following method. FIG. 7 shows a pattern in which loops are to be detected. Assume that points P and Q are the edges of the pattern. Using each point as a starting point, trace the edge while looking at the background to the right. Here, p is the trace path of point P, q is Q
It becomes a trace path of points. When each point P and Q traces the pattern and returns to the starting point, pay attention to the tracing route. If the direction of rotation is counterclockwise, the outer circumference is traced, and if it is clockwise, the inner circumference is traced. Become. Therefore, the number of routes traced in the clockwise direction is the number of loops of the desired pattern.

(Problems to be Solved by the Invention) However, in the conventional loop detection method described above, special hardware is required to trace the edge points of the pattern, and the scale of the hardware becomes large. There was a problem. The tracing algorithm is also complex.

It is an object of the present invention to eliminate the problems of the conventional technology such as the increase in hardware scale and the complexity of the tracing algorithm described above, and to provide a method that can detect the number of loops in a character/figure pattern using an easy-to-use algorithm. .

(Means for Solving the Problems) The present invention optically reads character figures on a form, obtains a black and white binary character figure pattern through photoelectric conversion, and performs character recognition based on the character figure pattern. In order to solve the above-mentioned problems of the prior art, the loop detection method in an optical character reading device is targeted, and the character figure pattern of interest is raster-scanned, and the coordinates of the white/black change point of the pattern during each main scan are calculated. By detecting the change point coordinates and comparing the change point coordinates with the change point coordinates from the previous scan, the continuity of strokes is managed using a table, and the number of loops in a character shape is detected based on the merging of the same strokes. This is what I did.

(Operation) In the present invention, the entire surface of the pattern to be recognized is scanned once, and coordinates at which the pattern changes from white to black and from black to white are detected in each main scan. The coordinates of the change point obtained in one main scan are then compared with the coordinates of the change point obtained in the previous main scan to check the continuity of the strokes and manage them using a table. Here, since the number of merging of the same stroke corresponds to the number of loops, the number of loops of the pattern is detected by counting the number of merging of strokes based on the table. The above-mentioned processing does not increase the size of the hardware and can be realized by a simple algorithm, thus solving the problems of the prior art.

(Example) An embodiment of the present invention will be described in detail below.

FIG. 1 is a block diagram showing this embodiment, and FIG. 2 is a diagram showing a working table that stores coordinate information of character patterns.

First, prior to explaining this embodiment, OCR to which this embodiment is applied will be explained with reference to FIG.

31 is a form in which characters to be read are written,
Characters to be read on the form 31 are illuminated by a light source 32, and the reflected light is focused by an optical lens 33, and the CCD
An image is formed on the sensor 34. The image formed on the CCD sensor 34 is photoelectrically converted, further converted into a digital signal by the A/D converter 35, and stored in the image memory 36. Here, in the image memory 36, by transporting the form 31 through the form transport system 37 in synchronization with the υCCD sensor 34, a character string is stored as one line of two-dimensional image data as shown in block 36. I can do things. Next, one line of character images stored in the image memory 36 will be recognized, but since character recognition is performed character by character, the character cutting unit 38 cuts out one pattern in the image memory 36 and creates a pattern. The data is stored in the memory 39. Then, for the pattern in the pattern memory 39, a feature extraction unit 40 detects character features such as loops, unevenness, and inclination of character lines, and registers the encoded character features in a feature 41. next,
In the character determination unit 42, a recognition dictionary 43 in which characteristics of all characters to be read are registered is a characteristic clothing 41.
The characters to be read are determined by comparing with the characters.

The above is the basic principle of OCR, and the present invention relates to loop detection of a character pattern, which is one of the important features among various features, in the feature extraction section 40.

Next, the present embodiment will be explained. In the block diagram of FIG. 1, 1 is a pattern memory corresponding to 39 in FIG. 3, in which pattern 2 to be recognized is stored.
The contents are the X coordinate indicated by the X counter 3 and Y counter 4,
It can be referenced arbitrarily by the Y coordinate.

To explain the loop detection procedure according to this embodiment, the first step is to operate the counter.The X counter 3 is counted up sequentially from O to m, and for each address of the X counter 3, the Y counter 4 is sequentially counted up from O to n. Count up. That is, here, the Y direction is the main scan,
This is raster scanning with the X direction as the sub-scanning. However, here, the Y direction is merely assumed to be the main scan, and if necessary, the X direction may be used as the main scan, or the scanning direction may be reversed.

In the process of raster scanning, when a character pattern black dot portion is reached, the coordinates of the start point and end point of the black portion are recorded for all black dot portions being scanned in the Y direction. Figure 4 shows the first
An enlarged view of the left end of character pattern 2 in the figure is shown.
In this example, the coordinates recorded first are Y, jY, +4. Then, these coordinates are registered in the work table shown in FIG. 2 according to the following rules.

(2) If the coordinates are not related to any already registered strokes, register the starting point of the Y coordinate of the extracted stroke in Y3 and the end point in Y in the table, and add a new stroke salt.

■ If this coordinate is related to a stroke that has already been registered, set the starting point of the Y coordinate of the extracted stroke to Y3 in the table and the ending point to Y. , and the stroke salt takes over the previously registered stroke salt.

Here, the relevance of strokes refers to the continuity between the strokes extracted in the previous main scan and the strokes extracted in the current main scan, and in the example of FIG. Since the stroke of YtSYt+2 is continuous with the stroke A of Yt+2 to Yt+4 of the scan of Xk, it is considered that they are related, and the stroke salt A is inherited. On the other hand, Yt+6 to Yt+ of scanning Xk+1
Since the stroke of 8 is not continuous with the stroke A in the scan of Xk, the stroke salt is newly set to B.

In accordance with the above two rules, the table shown in FIG. 2 is regenerated every time the sub-scanning is performed, and in this process, the following is found.

(a) Pattern generation--Apply a new stroke salt.

(b) Branching of pattern - 1 The same stroke name is additionally registered in the table.

(C) Merging of patterns-m-Identical or different stroke salts are registered as one common stroke.

(d) Disappearance of the pattern - m - Right edge of the pattern.

FIG. 6 shows the correspondence between the characteristic points of (a) to (d) above and the patterns. Of these, the point to pay attention to is C.
) is a confluence of patterns. A confluence of stroke names with different stroke names does not constitute a loop (point C in Figure 6), but a confluence of strokes with the same name means that when updating the table in Figure 2, it is not a loop. If the same stroke salt of is decreased by one due to merging, one loop has been detected there (point C in Figure 6). In this way, if the decrease in the same stroke salt due to merging is counted while scanning the entire pattern, the total number of loops included in this pattern will be counted.

(Effects of the Invention) As described above in detail, according to the present invention, a pattern of interest for which loop detection is to be performed is scanned once in a combined plane raster scan, and the white/black change point of the pattern during each main scan is performed. Loop detection can be performed simply by detecting coordinates and comparing these coordinates with the coordinates from the previous scan, so loop detection can be achieved with small-scale hardware, making it possible to construct a compact and highly accurate OCR. becomes.

Furthermore, since the method of the present invention can realize loop detection simply by scanning character patterns, it is
, φ can be easily applied to devices that require loop detection, such as pattern recognition devices.

[Brief explanation of the drawing]

Figure 1 is a block diagram showing an embodiment of the present invention, Figure 2 is a diagram showing a working table that stores coordinate information of character patterns, and Figure 3 is a block diagram of OCR to which the embodiment of the present invention is applied. , Figure 4 is a partially enlarged view of the pattern in Figure 1, Figure 5
The figure is an explanatory diagram of the continuity of the 0-1 pattern, FIG. 6 is a diagram showing the characteristic points of the pattern, and FIG. 7 is an explanatory diagram of the conventional loop detection method. ...X counter 4...Y counter

Claims (1)

[Scope of Claims] An optical character reading device that optically reads character figures on a form, obtains a black and white binary character figure pattern through photoelectric conversion, and performs character recognition based on the character figure pattern, The character/figure pattern of interest is raster-scanned, and the coordinates of the white/black change points of the pattern during each main scan are detected, and the change point coordinates are compared with the change point coordinates of the previous scan to determine the continuity of strokes. A loop detection method is characterized in that the number of loops in a character figure is detected based on the merging of the same strokes by managing the number of loops using a table.