JPS60217479A - Recognizer of character - Google Patents

Abstract

PURPOSE:To accelerate a character recognition processing by extracting strokes of a character pattern, then by extracting intersections and branch points of the character in accordance with the extracted connection relation and by connecting appropriately spaces between strokes comprising character intersections. CONSTITUTION:An unknown character 2 on a document 1 is read out by a character reader 3, its noises are removed in a pretreatment circuit 5 through an A/D convertor 4, and said character is stored in a picture memory. After a feature extraction circuit 6 encodes features (intersection, branch point, loop, etc.) of the character, it extracts them, while a dictionary collation circuit 7 collates them with a feature code of a standard pattern stored in a dictionary 8. When both of them are coincident, said circuit 7 outputs a discrimination label of the standard pattern having the feature code. If a candidate character cannot be specified by the feature code outputted by the feature extraction circuit 6, the dictionary collation circuit 7 trakes strokes comprising the candidate character, and collates the tracked result with a preset standard tracking pattern. When they are coincident, the discrimination label of the standard pattern having the standard tracking pattern is outputted.

Description

[Detailed Description of the Invention] <Technical Field of the Invention> The present invention optically reads unknown characters and converts them into black and white.
After converting into a value to obtain a character pattern, the character recognition device extracts the characteristics of the unknown character from the character pattern and compares these characteristics with the characteristics of a standard pattern stored in a dictionary in advance to identify the unknown character. Relatedly, in particular, the present invention provides a novel technique suitable for extracting intersections and branching points of characters that constitute character features.

<Background of the Invention> In general, a character recognition device extracts the presence or absence and number of character intersections, branch points, loops, etc. as character features, and uses these features to narrow down candidate characters for unknown characters, or finally performs detailed analysis. Unknown characters are recognized by checking the dictionary.

Conventionally, to extract character intersections and branching points among the character features mentioned above, a binary character pattern is scanned with an observation window of predetermined length and width bits, and the pattern within the window is determined as a predetermined reference pattern. 7, the corresponding black background bit is labeled, and finally, the combination state and distance relationship between these labels are checked to extract character intersections, etc. However, in this type of method, especially when handwritten characters are recognized, it is necessary to prepare a large number of reference patterns, and if the characters are deformed, the processing method for the labeled information becomes complicated and the processing time increases. There were problems such as the length of the

Further, in the detailed dictionary matching process described above, the strokes of the character pattern are traced, the tracing direction is converted into vector numbers, and each vector number is compared with the vector number of a standard pattern stored in advance. . Therefore, according to this matching method, for example, in stroke tracing of a character pattern P including character intersections as shown in FIG.
The strokes must be traced radially on all four sides from four directions toward the character intersection, which is disadvantageous in that it takes a lot of time to trace the strokes during the dictionary matching operation.

<Object of the Invention> The present invention extracts strokes of a character pattern, extracts character intersections and branching points from the connection relationships between each stroke, and then connects the strokes that make up the extracted character intersections as appropriate. The object of the present invention is to provide a character recognition device that solves the above problems.

<Structure and Effects of the Invention> In order to achieve the above object, the present invention traces the black background portion of the character pattern 7, specifically the boundary portion between the black background and the white background, in a predetermined direction to create a directional character pattern. After extracting the strokes, and then extracting the connection points between each stroke, trace the continuity of the connection points to extract character intersections, etc., and then connect the facing strokes among the strokes that make up the character intersections. I decided to do 2 things.

According to the present invention, there is no need for a reference pattern for extracting intersections, etc., as in the conventional method, and even if the character to be recognized is deformed, the process for extracting intersections, etc. is not affected in any way, making it easy and reliable. Character intersections and branching points can be extracted.Also, since we have adopted a method of connecting the strokes that face each other among the strokes that make up character intersections, character patterns with character intersections can also be extracted as shown in Figure 18. It has excellent effects such as being able to trace strokes in a cross pattern, performing dictionary matching in two tracing processes, and speeding up character recognition processing.

<Description of Examples> Fig. 1 shows the overall configuration of a character recognition device, in which an unknown character 2 on a form 1 is read by a character reading device 3, and the reading output is A.
The signal is converted into a time-series digital signal by the /D converter 4. The preprocessing circuit 5 performs noise removal and binarization processing on the digital signal, and stores the character pattern in an image memory (not shown).
to be memorized. The feature extraction circuit 6 encodes and extracts character features such as the presence or absence and number of intersections, end points, branch points, loops, etc. from the character pattern 7. The dictionary matching circuit 7 is
The feature code output by the feature extraction circuit 6 is compared with the feature code of the standard pattern stored in the dictionary 81, and when the two match, the identification label of the standard pattern having that feature code is output.

Furthermore, if the candidate character cannot be specified by the feature code output by the feature extraction circuit, the dictionary matching circuit 7 tracks the strokes that constitute the candidate character, and matches this tracking result with a standard tracking pattern prepared in advance. Then, when the two match, an identification label of the standard pattern having that standard tracking pattern is output.

Figure $2 shows a character pattern on the image memory and specifically shows a method for extracting features of unknown characters. The illustrated method focuses on the boundary line between the white background and the black background (indicated by a thick line in the diagram) that forms the outline of the character pattern, and determines the direction in which this boundary line extends in the four directions A and D shown in Figure 3. By tracking in the direction, we can obtain directional paired substrokes (A, ,
A, ) (B1.B2) (C1, C3) (Dl, B3)
(B2, B4) (C2, C4), and further extract strokes b, c, d, d', c' of the character pattern shown in Figure 4 based on the six pairs of substrokes. Thus, an approximate pattern P for feature extraction is obtained. In FIG. 4, the circles at both ends of each stroke indicate the end points, and as the end point information, the coordinates of the end points of each sub-stroke are stored in RAM (Randon Access Memory) shown in FIG.
y) are stored in a predetermined area m and d.

Next, for the end point coordinates of each sub-stroke, the coordinates in eight surrounding directions shown in FIG. 6 are scanned, and the connection relationship with other strokes is checked. When the connection between the strokes is confirmed, each piece of information about the connection points I to Ko is stored in a predetermined area m2 of the RAM shown in FIG. In Fig. 7, k means the right end point, L means the left end point, U means the upper end point, and D means the lower end point. For example, the information on the connection point is the right end point of the stroke end and the upper end of the stroke b. This means that the end point U is connected.

Next, the continuity of each connection point is checked based on the connection point information. For example, in the example shown in FIG.
, K2 are continuous sharing the left end point 1't of the stroke d', and similarly 2. K3 is stroke C'
3 at the right end point T2 and the connection point. K, is the right end point T3 of stroke d, and 4. K, are continuous and share the left end point T4 of the stroke C, so that 4
It is determined that the connection points 1 to 4 constitute the intersection of characters. In addition, as shown in Figure 9, 1 to 3 are connected to the three connection points.
When similar continuity is recognized in these connection points, 1 to
It is determined that the set of 3 constitutes a branching point of characters. Note that the above method according to the present invention is not applicable to the extraction of character branching points as shown in FIG. 10, where a character end point joins the middle of a stroke.

FIG. 11 shows a circuit for extracting character intersections and branch points I7 and converting the extracted intersection information.

In the figure, the image memory 10 stores preprocessed and binarized character patterns, and the stroke extraction circuit 11 extracts substrokes by tracking the character pattern 7, and then Extract the strokes of characters. The connection point extraction circuit 12 checks the connection relationship between each stroke and extracts connection points.

The stroke extraction circuit 11 extracts information on each stroke end point,
Further, the connection point extraction circuit 12 outputs information on each connection point,
The results are stored in RAM as shown in Figures 5 and 7 above.
13.

CPU (Central Processing)
Unisu 14 decodes the program in program memory 15, writes or reads data into image memory 10 and RAM 13, controls the operation of each extraction circuit 11, 120, extracts character intersections and branch points, and further Execute the conversion process of the extracted intersection information.

FIG. 12 shows the control flow of the intersection extraction operation.

First, in step 21, the CPU 14 loads connection point information, for example, stroke end point information rc-LJ of connection point 1 (3) shown in FIG. The scanned flag is set in the storage area of each connection point information, but in this case, the scanned flag is not set and is in the OFF state, so check step 22. The determination is No'', and in the next step 23, the scanned flag and the scan start flag indicating the start point of scanning are set to the on state.Then, in the next step 2
4, the connection partner stroke information [d-RJ is loaded, and after turning on the scanned flag for this information in step 25, in the next step 26, the continuity of the connection point, that is, the information ``including dRj'' is loaded. Search for other connection point information.In this case, the information of connection point lζ6 includes information 1
``d-RJ is included, so the determination of ``Is there continuity?'' in step 27 is YES''.In the information of 6 at this connection point, neither the scanning start flag nor the scanning completed flag is set. Since there is no scan, the result in both steps 28 and 29 is 'NO', and in the next step 30, the scanned flag is set, and the process returns to step 24, where the same steps are repeated.

Thus, starting from 3 at the connection point, 5°1 at the connection point, .
, and when the process returns to the first connection point I (3), the judgment in step 28 becomes ``YIi5'', the scan start flag is turned off in step 31, and the intersection point extraction counter is set in step 32. The content is incremented by 1. Then, in step 20, it is checked whether the scanned flag has been set in the information of all connection points, and when the determination in step 20 is ``YES'', the intersection extraction process is completed. .

The above example shows an example of extracting character intersection points, but if this is also used to extract character branching points, a counter is provided to count the connecting points, and the scanning start flag is turned on in step 23. When the counter is reset, the contents of the counter are incremented by 1 each time the scanning flow flag is turned on in steps 23, 25, and 30. Then, when the scanning start flag is turned off in step 31, if the counter content is "6", this is determined to be a branching point, and if the counter content is "8", this is determined to be an intersection point, Add 1 to the branch point extraction counter and the intersection point extraction counter.

Furthermore, in the above control flow, when the determination of "Is there continuity?" in step 27 is ``YES'', that is, when a character intersection is extracted, the identification label of the connection point constituting the intersection is Predetermined area m3 of RAM 13 (first
(shown in Figure 4) in the order of continuity check (K3. K6.
K, , K4).

FIG. 13 shows a control flow of an operation for connecting strokes that face each other among strokes that constitute an intersection, and this operation is executed at step 32 in FIG. 12.

First, in step 41, cpul 4 refers to area m3 shown in FIG. 14 and selects the top two connection points as 3. Then, the stroke end point information [d-Rj shared by is extracted from the area m2 shown in FIG.

Next, in step 42, cpul 4 similarly refers to area 1113 and connects the lower two connection points 1 (5°K4
The stroke end point information Nrd'-LJ that is shared by the area m
Extract from 2. Then, these extracted stroke end point information [d-RJ and [d'-LJ are the end points IN of the facing strokes d, d/, and in the next step 43, these end point information [d-RJ, rd '-L
J is newly stored in area m2 as information of 3 at the connection point (see FIG. 15).

Next, C;I'U14 adds 3 to the connection point in step 44.
．． K6. K5. Extract the stroke end point information [c-LJ, [c'-RJ] other than the information ``d-Rj, rd・-Lj'' from all the stroke end point information constituting K, and newly connect these pieces of information. Point K.

In step 45, end point information 3 to 6 is stored in the memory m2 as information on the previously stored connection points in the area m.
Delete from 2.

FIG. 15 shows the RAM 13 after the above series of processing is completed.
FIG. 16 shows the memory configuration of area m2 in area m2, and FIG.
An image of an approximate pattern P expressed based on new connection point information in the figure is shown.

[Brief explanation of drawings]

Fig. 1 is a block diagram showing the overall configuration of the character recognition device, Fig. 2 is an explanatory drawing of character patterns showing the method for extracting character features, Fig. 3 is an explanatory drawing showing the tracking direction, and Fig. 4 is an illustration of stroke extraction. 5 is an explanatory diagram showing the information storage state of each stroke end point, FIG. 6 is an explanatory diagram showing the scanning direction, FIG. 7 is an explanatory diagram showing the storage state of connection point information, 8 to 10 are explanatory diagrams showing examples of stroke-extracted approximate patterns, FIG. 11 is a block diagram showing a configuration example of a character intersection extraction circuit, and FIG. 12 is a flowchart showing control operations for intersection extraction processing. , FIG. 13 is a flowchart showing the control operation of the process of connecting facing strokes among the strokes that make up the intersection, FIG. 14 is an explanatory diagram showing the storage state of identification labels of the connection points that make up the intersection, and FIG. 15 is an explanatory diagram showing the storage state of connection risk information after the end of the flow shown in FIG. 13, FIG. 16 is an explanatory diagram showing an approximate pattern expressed based on the connection point information in FIG. FIG. 18 is an explanatory diagram showing stroke tracing at the intersection point f in the conventional method, and FIG. 18 is an explanatory diagram showing stroke tracing at the intersection point according to the present invention. Patent Applicant Tateishi Electric Co., Ltd. -2A 2 Self-mouth β -A51 Poor 6 Diagram I92 Fang// Diagram/: 37 Hai/mu Diagram 76 Diagram Min.lq Zuga 75

Claims (1)

[Claims] After reading unknown characters and converting them into black and white binarization to obtain a character pattern, the device extracts the features of the unknown characters and compares them with standard patterns to recognize the unknown characters.The black background part of the character pattern is tracked to obtain directionality. Character pattern straw
a means for extracting connection points between each extracted stroke; a means for extracting character intersections and branching points by Kokichi for tracing the continuity of connection points; and a means for extracting connection points between each extracted stroke. A character recognition device comprising a means for connecting facing strokes among the strokes.