JPS61123987A - Character recognizing device - Google Patents

Abstract

PURPOSE:To select and specify one character out of plural candidate characters by scanning a character pattern in a prescribed direction from a prescribed position of the pattern and detecting whether or not a scanning point has reached the prescribed part of the character pattern. CONSTITUTION:A character feature extraction circuit 10 extracts features of the character pattern, and houses them in a RAM 12. A collating part squeezes them into a pair of characters corresponding to the inputted character pattern such as 'E' and 'Q'. Then the coordinates of a specified character end point 41, for instance, are inputted from the RAM 12. Information (arrow P) concerning a specified direction is inputted from a dictionary 7, and such a character pattern scanning is carried out as to transfer by one picture element in the specified direction P starting from the endpoint 41. When the reaching point of a scanning point is an observation frame, a character other than 'F' is selected, and when the reaching point is in black and of separated coupling, 'E' is selected out of candidate characters, thereby outputting them to an indicating part. Thanks to this, a character, which cannot be discriminated by a direction or a number of a character stroke can be securely recognized.

Description

DETAILED DESCRIPTION OF THE INVENTION [Technical Field of the Invention] The present invention extracts geometric features of character patterns of unknown characters and compares the features with standard patterns stored in a dictionary section.

The present invention relates to a character recognition device that recognizes unknown characters.

<Summary of the Invention> The character recognition device of the present invention scans a character pattern from a predetermined position on the pattern in a predetermined direction, detects whether a scanning point has reached a predetermined portion of the character pattern, and thereby This method selects and specifies one character from among multiple candidate characters, making it possible to reliably recognize characters that cannot be determined based on the direction or number of strokes of a single character.

<Background of the invention> Generally, in a character recognition device, as shown in FIG.
The characters, symbols, and figures displayed above (hereinafter collectively referred to as "characters") are optically read by the reading unit 2, and the character pattern is converted into black and white and stored in the image memory 4. . Next, the preprocessing circuit 3 inputs the data in the image memory 4 and performs so-called preprocessing such as noise removal, smoothing, and thinning, and then the feature extraction circuit 5 inputs the preprocessed characters. Geometric features of the pattern are extracted from the pattern.

The dictionary matching circuit 6 matches the features extracted by the feature extraction circuit 5 with the standard patterns of each character stored in advance in the dictionary 7, and determines how many characters are considered to correspond to the read character pattern. character (“candidate character”)
), one character is identified from among the candidate characters using detailed information on character characteristics, and the character is output as a result on a display unit (not shown), etc. It is designed to be given to

Of these, the dictionary matching circuit 6 narrows down the list to a plurality of candidate characters corresponding to the input character pattern, using data on the character features extracted by the feature extraction circuit 5, such as loops, one branch at an intersection, and end points. First, one of them is selected and specified using character stroke information and the like. For example, when a character pattern as shown in FIG. 12(A) is input in the kana character recognition mode, the character pattern shown in FIG.
As shown in B), the character features of this character pattern include 6 connected components, 1 intersection (indicated by C in the figure), and 6 character endpoints (indicated by white circles in the figure). As a result, katakana characters "o" and "chi" are candidate characters based on this information. However, when we try to identify one of these candidate characters using the character stroke information, as shown in Figure 13, the two candidate characters "o" and "chi" both have the same stroke information. Therefore, identification becomes impossible. For this reason, in the past, a method has been adopted in which one character is specified from these candidate characters using roughly different character feature data, which has the disadvantage of complicating the process.

<Objective of the Invention> The present invention is intended to solve the above-mentioned drawbacks, and to provide a method for relatively easily and accurately character recognition even in cases where one character among candidate characters cannot be identified using only character stroke information. The object of the present invention is to provide a character recognition device that can perform the following functions.

<Structure and Effects of the Invention> In order to achieve the above-mentioned object, the character recognition device of the present invention, when a plurality of candidate characters are specified, determines predetermined position information on a character pattern according to a set of these candidate characters. reading means for reading out predetermined scanning direction information from the dictionary section; scanning means for scanning the character pattern from a predetermined position on the character pattern in a predetermined scanning direction based on the succession information; It has been decided to provide means for detecting whether or not the character pattern has been reached and selecting and specifying a character corresponding to the character pattern from among a plurality of candidate characters.

According to this invention, even in cases where a single character among candidate characters cannot be identified using character features such as loops, intersections, and branch points or character stroke information, it is possible to identify the character relatively easily and accurately. This has the effect of improving character recognition accuracy and improving performance.

<Description of Embodiments> FIG. 2 shows the main parts of a character recognition device which is an embodiment of the present invention. In the figure, a preprocessing circuit 3, an image (image memory 4)
The dictionary 7 has a configuration and function similar to that shown in FIG. The labeling circuit 8 labels the read character patterns to be described later, thereby forming the stroke extraction circuit 9 and the character feature point extraction circuit 1.
Assists the movement of 0. The stroke extraction circuit 9 is for extracting strokes from character patterns, while the character feature point extraction circuit 10 is for extracting various feature points of character patterns. CPU (Ce
ntralProcessing Unit) 11
The CPU II controls each part of this device, and operations according to the flow described later are also performed mainly by the CPU II. RAM (Rondem Access
Memory 12 is a memory that stores programs for the CPU 11 and various data for work. The rest of the configuration is almost the same as the device shown in FIG.

Next, the operation of the apparatus having the above configuration will be explained. Here, the pattern shown in FIG. 12(A) is assumed as the input character pattern. First, the optically read character pattern is stored in the image memory 4, and then the data of the character pattern is subjected to preprocessing such as noise removal and smoothing in the preprocessing circuit 3. Next, labeling circuit 8
The pixels constituting the character pattern are labeled by this method. As shown in Figure 3, this labeling is performed on pixels that form the black background part (character part) of the character pattern and are located at the border with the white background part. A bow-shaped mark is attached to the adjacent side. For example, since the pixel 21 in FIG. 3 is adjacent to a white background on its upper side, an arcuate mark is placed on its upper side. Hereinafter, these labels will be distinguished using the symbols (u), (d), (1), and (r), depending on whether these bow-shaped marks are attached to the top, bottom, left, or right side of the pixel. call. Therefore, for example, pixel 21 has a bow mark on its upper side, so the label (u
), and the pixel 22 is labeled (u, r) because there are arcuate marks on the top and right sides. In this way, the pixels in Figure 3 are labeled, and the (x, y) coordinates of each pixel and its label are stored in a chain called NC0T in the RAM 12, as shown in Figure 4. .

In this storage, first, scanning is performed in the direction shown by the arrow 20 in FIG. 3, and the character pattern is traced counterclockwise from the pixel of the label (u, r) that is reached first.

Here, counterclockwise tracking refers to the label of the pixel where the tracking point is located at that point (5th
In the figure.

Focus on Al~D1 (shaded arcuate mark).

In the case of label A1, whether or not a label having the direction and position A2-A4 exists among the labels adjacent to the label is sequentially detected in the order of this person 2 to A4, and if it exists, refers to moving the tracking point to the coordinate position of the detected label. For example, pixel 23 in FIG.
If you pay attention to the A1 label on the bottom side of the label,
Since this label is not adjacent to the A2 type label and is adjacent to the A3 type label on the right side, the tracking point is moved to the pixel 24 of this label A3. Fifth
The same applies to figures (B) to (D), 82 to B4, C2 to
Label detection and tracking are performed in each order of C4, D2' to D4.

Through this tracking, when pixels located at the outer boundary of the character pattern are sequentially detected, the position coordinates and label type are stored in the memory area NC0T shown in FIG.
It will be stored above. On the other hand, when a tracking point goes around one connected part of a character pattern, it returns to the original tracking starting point, forming a closed loop. 4. Provide an index labeled "stopper" in Figure 4. Thereafter, the next connected portion is tracked and the same process is repeated. Therefore, the number of "stoppers" in Figure 4 matches the number of connections in the character pattern, and data stored between one stopper and another means data regarding the same connecting part. It turns out.

After labeling and data storage are completed in this way,
Next, the stroke extraction circuit 9 of FIG. 2 traces the continuity in eight directions as shown in FIG. 6 in the pattern of FIG. 3, and generates an approximate character stroke as shown in FIG. Extract patterns. Then, using the pattern in FIG. 7 and the labeled character pattern in FIG. 3, the character feature point extraction circuit 10 extracts the geometric features of the character pattern. This feature also includes the coordinates of pixels corresponding to the end points (white circles in Figure 7) of each stroke forming one character pattern, and this end point data is stored in the CHT M chain in the RAM 12 shown in Figure 8. Stored in acid.

Next, a collation unit (not shown) corresponding to the dictionary collation circuit 6 in FIG. 11 compares data regarding these feature points and
The candidate characters are compared with standard patterns previously stored in the dictionary 7, and the candidate characters corresponding to the input character pattern shown in FIG. 12 are narrowed down to a character set consisting of "o" and "chi".

Then, the character stroke in Figure 7 is compared with the character stroke information of the candidate character shown in Figure 13, but in this case, as mentioned above, this character stroke information alone cannot
It is not possible to determine whether it is "o" or "chi", and the identification will be made by executing the process shown in the flowchart of FIG. 1 below.

First, in step 31, the coordinates of a predetermined position on the character pattern are loaded, corresponding to the character set consisting of "o" and "chi". Information on this position is set in the dictionary 7, and for example, as shown in FIG. 7, the upper end point 41 of the upward-sloping stroke can be used (hereinafter referred to as the "designated character end point"). Specifically, by this step 31, the locus 11 (7, 12) of the corresponding end point existing in the area CHT to (2) in FIG. 8 is loaded. In the next step 32, information specifying the scanning direction (hereinafter referred to as "designated direction") is similarly loaded from the dictionary 7. In the illustrated example, this specified direction is.

This is the direction indicated by the dotted line arrow P in FIG. Next step 3
In step 3, the character pattern is scanned with the designated character end point as the starting point and the designated direction as the scanning direction. This scanning is performed by moving the scanning point by one pixel. In the above example, the scanning point first moves from the end point (7, 12) to (8, 12) in FIG.

In step 34, it is determined whether the +mi element of the coordinates of the scan point reached is a black background (character portion). In the case of the above coordinates (8, 12), that pixel is not a black background, so
Move on to step 35.

Next, it is determined whether the scanning point has reached observation frame F in FIG. When it reaches observation frame r, it means that it has reached observation frame F without reaching any other character part of the one-character pattern, but if it is "o", it should reach connected part 25. Therefore, it is determined that this character pattern is not the character "O" as shown in FIG. 12(B), and the process moves to the main routine or the other process as an error return.

If it is determined in step 35 that the observation frame is not F, the process returns to step 33 to continue scanning.

If it is determined that the background is black in step 34, step 3
6, it is determined whether the black background part belongs to a connected part different from the connected part to which the designated character endpoint belongs. This can be easily determined by referring to the stored data since the coordinate storage pattern shown in FIG. 4 is arranged for each directly connected portion. If it is not a separate connection, it means that the same connection has been reached, but
In this case, the process returns to step 33 and the rough running continues.

In step 36, if the character pattern is different concatenation and the same cut f is performed, it turns out that this character pattern is "o" (such a result does not occur in "chi J"), so "o" is selected from the candidate characters. This character is output to a display unit, etc., and the process is completed.

This kind of judgment is also possible for other character sets, such as distinguishing between "su" and "ma", "u" and "wo", as shown in Figures 9 and 10, respectively. In this case, the endpoints 42, 43 and 4 are used as the specified character endpoints.
4, 45, and the dotted arrow Q,
The directions indicated by R may be adopted.

[Brief explanation of drawings]

Fig. 1 is a flowchart showing the main parts of the processing in the embodiment of this invention, Fig. 2 is a block diagram showing the main parts of the character recognition device which is the embodiment of the invention, and Fig. 3 is the labeling. Figure 4 is a diagram showing an example of storing labeled data, Figure 5 is a diagram to explain counterclockwise tracking, and Figure 6 is a diagram to explain approximate strokes with directionality. , Figure 7 is a diagram showing character strokes and scanning directions, Figure 8 is a diagram showing an example of storing end point coordinates, Figures 9 and 10 are diagrams showing scanning directions, and Figure 11 is a diagram showing character recognition. FIG. 12 is a block diagram showing the general configuration of the device; FIG. 12 is a diagram showing input character patterns and their character strokes; FIG.
The figure is a diagram showing character stroke information of candidate characters. 4...m@memory 5...Feature extraction circuit 6.
...Dictionary matching circuit 7...Dictionary 8...Labeling circuit 9...Stroke extraction circuit 10...Character feature point extraction circuit 11...CPU 12...λAM' Patent applicant Tateishi Electric Co., Ltd. ÷Y no Katana δ Tsukujiku Sunia・r7ruQxri no Kimaro 3&Shizu flowchart masu//K Wa? J ``hi #do bite 11shi'' - 1 cor 9 - 匈ha 1 to luck 1 (A1 column
Zui Shimisokamet, IΩ

Claims (1)

[Claims] A character pattern of an unknown character is read, the pattern is converted into black and white to obtain an input image, and after a feature extraction section extracts the geometric features of the input image, a matching section The character recognition device recognizes the unknown character by comparing the output of the feature extracting unit with a standard pattern stored in a dictionary unit, and when the matching unit narrows down a plurality of candidate characters, means for reading out predetermined position information and predetermined scanning direction information of a character pattern from the dictionary section according to the set of characters; and means for scanning the character pattern from the predetermined position on the character pattern in a predetermined scanning direction based on the readout information. A character recognition device comprising: means for detecting whether a scanning point has reached a predetermined portion of a character pattern and selecting and specifying a character corresponding to the character pattern from among the plurality of candidate characters.