JPH028352B2 - - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION [Field of the Invention] The present invention relates to a character recognition device that recognizes characters such as Chinese characters that are composed of many straight line segments (hereinafter referred to as strokes).

[Configuration and operation of conventional technology]

Conventionally known character recognition devices of this type are equipped with a recognition dictionary in which a reference midpoint pattern, which is a midpoint pattern of the stroke of a reference character, is stored in its original format. It was said to be a composition. For this reason, it was necessary to make the memory capacity of the recognition dictionary extremely large.
FIG. 1 shows, for example, a stroke midpoint pattern 1 extracted from the character "gold" by stroke direction,
It is a figure which shows the reference|standard midpoint pattern 2 corresponding to this. The reference midpoint pattern 2 shown in FIG. 1 is created by normalizing the frequency of occurrence of the midpoint of the stroke of the reference character, and the size of the reference midpoint pattern 2 is 16×16. Each point can take on four values from 0 to 3. Therefore, 1
The memory capacity of the recognition dictionary required to memorize the reference midpoint pattern 2 of characters is 16 x 16 x 2 bits x 4 types = 2048 bits = 256 bytes, and for example, for 1000 character types it will reach 256 kilobytes. . As described above, since the memory capacity of the conventional recognition device described above becomes very large, the reference midpoint pattern 2 is compressed using a well-known technique such as the run-length method, and stored in a recognition dictionary. A method for restoring and using this is known.

[Disadvantages of conventional technology]

Since the conventional character recognition device is configured as described above, as described above, in a character recognition device equipped with a recognition dictionary, the reference midpoint pattern 2, which is the midpoint pattern of the stroke of the reference character, is stored in the recognition dictionary.
Since the recognition dictionary is stored in its original format, it is generally necessary to make the memory capacity of the recognition dictionary very large, which has the drawback of making the device large and expensive. On the other hand, as a method to solve this drawback, the reference midpoint pattern 2 is compressed using a well-known technique such as the run length method, stored in a recognition dictionary, and then restored and used during recognition. This method is known, but when using this type of method to compress the reference midpoint pattern 2, new drawbacks arise such as the large amount of processing time required for restoration during recognition. However, even with this method, there was a problem in that it could not fundamentally solve the above-mentioned drawbacks.

[Object of the present invention]

The present invention has been made in order to eliminate the drawbacks of the conventional ones as described above, and includes a thinning means for obtaining a thinning pattern from an input character pattern, and a method for extracting strokes and their midpoints from the thinning pattern and inputting the extracted strokes and their midpoints. a stroke midpoint pattern extraction means for determining a midpoint pattern of a stroke of a character; a recognition dictionary configured with a reference midpoint pattern that is a midpoint distribution of strokes of a plurality of reference characters; determining means for determining an input character based on the degree of similarity with the reference midpoint pattern; To provide a character recognition device in which only row or column information in which is not zero is stored in the recognition dictionary, thereby significantly reducing the memory capacity of the recognition dictionary and suppressing an increase in restoration time during recognition. It is intended to.

[Configuration of embodiment of the present invention]

Hereinafter, one embodiment of the present invention will be described with reference to the drawings. FIG. 2 is a block diagram showing a character recognition device according to an embodiment of the present invention. In FIG. 1, 3 is a form, 4 is a scanning means for scanning the characters recorded on the form 3, 5 is a thinning means for obtaining a thinning pattern from an input character pattern, and 6 is a stroke and its contents from the thinning pattern. Stroke midpoint pattern extraction means extracting points to obtain a midpoint pattern of the stroke of an input character; 7 is a recognition dictionary composed of standard midpoint patterns that are midpoint distributions of strokes of a plurality of standard characters; 8 is an input character This determining means determines an input character based on the degree of similarity between the midpoint pattern of the stroke and the reference midpoint pattern.

FIG. 3 is an explanatory diagram showing a method of compressing a reference midpoint pattern used in the character recognition device of FIG. 2.
In FIG. 3, 2a is a part (horizontal stroke) of the reference midpoint pattern 2 shown in FIG. 1, 9 is horizontal projection information, and 10 is vertical projection information. here,
To simplify the explanation, an X-axis and a Y-axis are introduced and shown with the origin at the upper left corner of FIG. 3.

FIG. 4 is a diagram showing the format of the reference midpoint pattern stored in the recognition dictionary in the character recognition device of FIG. 2. In FIG. 4, from the top, a matrix identifier 11 indicating row information or column information, and horizontal projection information 9.
and information from the 6th column to the 10th column of the reference midpoint pattern 2 are shown, respectively.

[Operation of the embodiment of the present invention]

Next, the operation of a character recognition device which is an embodiment of the present invention will be explained using the above-mentioned FIGS. 1 to 4. First, the characters recorded on the form 3 shown in FIG. 2 are scanned by the scanning means 4 and converted into a character pattern for each character, and then converted into a thinning pattern by the thinning means 5. Next, from this thinning pattern, the stroke midpoint pattern extracting means 6 extracts, for example, a stroke midpoint pattern 1 for each stroke direction shown in FIG. 1. Midpoint pattern 1 of this stroke extracts feature points such as end points, branch points, and bending points from the thinning pattern, finds basic line segments connecting these, and connects feature points other than the end points from this basic line. It is obtained by extracting strokes by connecting basic line segments whose connected directions are approximately the same, and then finding the midpoint of these strokes. Details of such an extraction method are disclosed in Japanese Patent Application No. 1989-89203 previously filed by the present applicant. The recognition dictionary 7 includes horizontal projection information 9 or vertical projection information 10 of the reference midpoint pattern 2 shown in FIG. row or column information is stored. Further, the determining means 8 calculates the degree of similarity between the stroke midpoint pattern 1 of the input character obtained by the stroke midpoint pattern extraction means 6 and the reference midpoint pattern 2 compressed and stored in the recognition dictionary 7. The character with the highest degree of similarity is determined as the recognition result and the input character is determined. Here, the degree of similarity can be determined by a well-known technique by adding the values on the reference midpoint pattern 2 at the midpoint positions of the strokes of the input characters.

Now, as can be easily understood from the disclosure of the drawing, the part 2a of the reference midpoint pattern 2 shown in FIG. 3 is mostly occupied by an area where the value of the reference midpoint pattern 2 is zero. It can be easily imagined that the memory capacity of the recognition dictionary 7 would be reduced if these areas were removed. However, as explained in the above-mentioned prior art, if the reference midpoint pattern 2 is compressed using the run-length method and stored in the recognition dictionary 7, a large amount of processing time is required to restore it during recognition. I will do it. In fact, in the similarity calculation method applied to the present invention, it is sufficient to know the value of the reference midpoint pattern 2 at the midpoint position of the stroke of the input character, and the entire reference midpoint pattern 2 is not required. The conventional technology for compressing and restoring the reference midpoint pattern 2 is particularly disadvantageous. Therefore, in the present invention, although the compression rate is not so high, in order to simplify the restoration process, a method is used in which only row or column information for which the value in the reference midpoint pattern 2 is non-zero is stored in the recognition dictionary 7. ing. Here, whether it is row or column information depends on the contents of the reference midpoint pattern 2. For example, in the reference midpoint pattern 2 shown in FIG. 3, there are 10 rows as row information and 5 columns as column information, so the column information is stored in the recognition dictionary 7. At the same time, horizontal projection information 9 indicating the stored column number and information indicating whether it is row information or column information are also required. In this case, the memory capacity of the recognition dictionary 7 is (5 columns x 16 x 2 bits + 16 + 1) = 177 bits as shown in Fig. 16 which is the memory capacity when stored in the format
Compared to ×16 × 2 bits = 512 bits, it is reduced to about 1/3. The content of standard midpoint pattern 2 differs depending on the character, and some have quite complex shapes, but computer simulations show that within the range of educational kanji, the average is about 1/2. This has been confirmed, and the effects achieved are significantly greater. Also, as shown in FIGS. 3 and 4, for example, the position P(X
=7, Y=6) on the reference midpoint pattern 2 can also be obtained by a simple algorithm as described below.

(1) Check the matrix identifier 11, and if column information is stored, exchange the values of X and Y. As a result, P(X=7, Y=6) becomes P(X=6, Y=
7).

(2) Y=7 in horizontal (vertical) projection information 9 (10)
It is checked whether the value at the position is significant (value 1), and if it is not significant, it is determined that the value on the reference midpoint pattern 2 is zero. As described above, since the values on the reference midpoint pattern 2 are zero at most positions, the processing ends at this stage, and the average processing time required for restoration is satisfactorily short. In this example, it becomes significant as indicated by the △ mark in FIG. 4, and the following processing is executed.

(3) Examine the horizontal (vertical) projection information 9 (10) from the left side and find out what significant value the significant value (value 1) at the position of Y=7 is. In this example, as shown in FIGS. 3 and 4, it is the second position, and it is found that the value of the position P is stored in the second column (row) information.

(4) Finally, if we find the X=6th value in this column (row) information, this is the original position P (X=7, Y
=6) is the value of the reference midpoint pattern 2.

As described above, in the present invention, by using the matrix identifier 11 described above, it is possible to easily obtain the value of a corresponding position on the reference midpoint pattern 2 at any position using a unified algorithm and in a short time.

[Other embodiments of the present invention]

Although the above embodiment describes the case of a stroke midpoint pattern in which strokes are divided into directions, the present invention is not limited to this.
The present invention is also fully applicable to the case of a stroke midpoint pattern in which the strokes are divided into lengths, and the same effects as in the above embodiment can be achieved.

[Effects of the present invention]

As described above, according to the character recognition device according to the present invention, the horizontal or vertical projection information of the reference midpoint pattern, which is the midpoint distribution of strokes of a plurality of reference characters, and the line of the reference midpoint pattern corresponding to the horizontal or vertical projection information Alternatively, among the column information, only the row or column information for which the projection information is significant is stored in a recognition dictionary, and the determining means restores the reference midpoint pattern from the recognition dictionary and matches the midpoint pattern of the stroke of the input character. Since the system is configured to calculate the similarity from the reference midpoint pattern, the memory capacity of the recognition dictionary can be significantly reduced, and the time required to calculate the similarity can be suppressed to a small increase. As a result, it is possible to construct an extremely small and low-cost character recognition device, which is an excellent effect.

[Brief explanation of drawings]

FIG. 1 shows, for example, the midpoint patterns of strokes extracted from the character "kin" by stroke direction,
A diagram showing the reference midpoint pattern corresponding to this, the second
FIG. 3 is a block diagram showing a character recognition device according to an embodiment of the present invention; FIG. 3 is an explanatory diagram showing a method for compressing a reference midpoint pattern used in the character recognition device shown in FIG. 2;
FIG. 4 is a diagram showing the format of the reference midpoint pattern stored in the character recognition device of FIG. 2. 1... Stroke midpoint pattern, 2... Reference midpoint pattern, 3... Form, 4... Scanning means, 5
...Thinning means, 6... Stroke midpoint pattern extraction means, 7... Recognition dictionary, 8... Determination means, 9
...Horizontal projection information, 10...Vertical projection information, 11
...Matrix identifier. In addition, in the figure, the same reference numerals are the same,
or a corresponding portion.

Claims (1)

[Claims] 1. In a character recognition device that recognizes characters recorded on a form, etc., there is a thinning means that obtains a thinning pattern from an input character pattern, and a thinning means that extracts straight line segments (strokes) and their midpoints from the thinning pattern and inputs them. a stroke midpoint pattern extraction means for determining a midpoint pattern of a stroke of a character; a recognition dictionary configured with a reference midpoint pattern that is a midpoint distribution of strokes of a plurality of reference characters; determining means for determining an input character based on the degree of similarity with the reference midpoint pattern, and the recognition dictionary includes horizontal or vertical projection information of the reference midpoint pattern and a line of the reference midpoint pattern corresponding to the horizontal or vertical projection information. Alternatively, among the column information, only the row or column information for which the projection information is significant is stored, and the determining means restores the reference midpoint pattern from the recognition dictionary, and combines the midpoint pattern of the stroke of the input character with the reference A character recognition device characterized by calculating similarity from a midpoint pattern.