JP3250317B2 - Character recognition device - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a character recognition device for recognizing handwritten characters and printed characters. In recent years, with the spread of information processing devices such as computers, a character recognition technology for reading handwritten characters or printed characters written on a form or the like has been developed as an input means to these devices. Since deformation such as distortion and inclination is larger than that of printed characters, a highly accurate recognition technique is required.

[0002]

2. Description of the Related Art A conventional character recognition apparatus will be described below with reference to the drawings. FIG. 10 is a block diagram of a conventional character recognition device. 10, reference numeral 100 denotes a character image input unit, 101 denotes a segment extraction unit, 102 denotes a feature extraction unit, 103 denotes a candidate determination unit, and 104 denotes a dictionary.

In a conventional character recognition apparatus, first, in a character image input unit 100, a handwritten character or a printed character written in a specific position such as a form is read as an image, converted into a black and white binary image, and segmented. Extraction unit 101
In, for example, a contour of the binary image is extracted as a character segment from the converted binary image. Then, in the feature extracting unit 102, the geometric shape information of the character image is extracted as the extracted feature amount of the contour line, and the candidate determining unit 103 given the extracted feature amount sets the feature amount in advance. The distance from the feature amount of the dictionary segment stored in the dictionary 104 is calculated, and the dictionary having the closest distance is output as a recognition result.

FIG. 11 is a diagram illustrating character recognition based on the geometric shape of a character image. As shown in FIG. 11A, the outline of the character image “2” has, for example, one concave portion (indicated by an arrow in the figure) on the left side and one on the right side as a geometric shape. In addition, the outline of the character image “3” shown in FIG. 11B has two concave portions on the left side and one concave portion on the right side.

[0005] As described above, using a geometrical shape different for each character image such as unevenness in the outline of the character image as a feature amount for identifying the character image, the character is recognized by paying attention to a shape unique to each category. This is particularly useful for recognizing characters having locally different shapes.

[0006]

However, when specifying which part of the dictionary segment stored in the dictionary corresponds to the local shape of the character segment of the character to be recognized, particularly large deformations occur. If characters are not efficiently associated with each other, there is a problem that the recognition rate is reduced and the time required for the recognition process is reduced.

For example, as shown in FIG. 11C, the dictionary segment has one concave portion on the left side, whereas the concave portion is formed as shown in the character image to be recognized shown in FIG. 11D. If there are two on the left side, it is necessary to process which concave portion corresponds to the concave portion of the dictionary segment. In view of the above, an object of the present invention is to provide a character recognition device capable of recognizing a character by accurately associating a character segment with a dictionary segment.

[0008]

FIG. 1 is a block diagram showing the principle of the present invention. In the present invention, as shown in FIG. 1, a means for solving the above problem stores a segment of a character extracted from a character image as a dictionary segment, and stores a dictionary at a reference point taken on the dictionary segment. the characteristics of the segments and dictionary 10 for storing a dictionary feature information of the character image, to take a point on the character segment extracted from character image to be recognized, a correspondence between the points on the dictionary segment And angle change
Correspondence Determination Means 1 by DP Matching Method Using Cumulative Value of
1 and a point of a character segment corresponding to a reference point on the dictionary segment is set as a check point based on the correspondence obtained by the correspondence determination means 11, and the dictionary segment at the reference point Test point detecting means 12 for extracting the characteristic information of the character segment at the test point corresponding to the dictionary characteristic information of the above, and the characteristic information of the character segment extracted by the test point detecting means 12 It is provided with candidate determining means 13 for determining a recognition candidate of the character image to be recognized by comparing with dictionary feature information of a reference point.

[0009]

According to the present invention, based on the above configuration, it is determined whether each point on the dictionary segment corresponds to a point on the character segment of the character image to be recognized. Inspection points on the character segments of the character image to be recognized corresponding to the reference points are determined.

Then, at the obtained inspection point, the characteristic information of the character segment corresponding to the characteristic information at the reference point on the dictionary segment is calculated and compared with the characteristic information of the dictionary segment. It is determined whether the character image to be recognized is a recognition candidate. Therefore, the feature information at the point having the characteristic recognition information on the dictionary segment to be noted for recognition for each category and the feature information on the character segment of the character image to be recognized corresponding to this point are Characters can be easily recognized by efficient comparison.

[0011]

Embodiments of the present invention will be described below with reference to the drawings. FIG. 2 is a block diagram of a character recognition device according to an embodiment of the present invention. 2, reference numeral 20 denotes a character image input unit, 21 denotes a segment extraction unit, 22 denotes a correspondence determination unit, 23 denotes an inspection point detection unit, 24 denotes a candidate determination unit, and 25 denotes a dictionary.

The character image input section 20 is for inputting characters written on a form or the like as a black and white binary two-dimensional pattern character image, and converts the characters written on the form or the like into a two-dimensional pattern image. The scanner 201 includes a scanner 201 for inputting an image, and a binarizing unit 202 for converting a character image input from the scanner 201 into a black and white binary image.

The segment extracting section 21 is for extracting a character segment from the binary image of the character obtained by the character image input section 20, and includes a contour line extracting section 211, an end point detecting section 212, and a segment dividing section 213. It is configured. FIG. 4 is a diagram illustrating extraction of a character segment. FIG. 4A is a binary image of a character image, and a hatched portion indicates a character portion represented by black pixels.
FIG. 4B is a contour line extracted from the binary image shown in FIG. 4A, and the dotted line portion shows the original binary image, and FIG. 3B shows a state in which the contour shown in FIG. 3B is divided into segments (S1 and S2) and end points (T1 and T2).

The contour line extracting section 211 extracts a contour line of a character portion represented by a black pixel on the binary image of the character obtained by the binarizing section 202 of the character image input section 20, and detects an end point detecting section. At 212, an end point corresponding to a part where writing of a character starts and ends of writing is detected from the extracted outline. FIG. 5 is a diagram illustrating the detection of an end point.

As shown in FIG. 5, an end point is a place where the contour changes rapidly, and three points (A, B, C) are placed on the contour (S) at regular intervals, and the three points are defined. The value is determined as a contour line region in which the value of the angle (θ) formed with the middle point (A) as the vertex is equal to or less than a predetermined value. Segment dividing section 21
In step 3, the outline is divided into segments based on the detected end points, and the segments extracted for each character image are stored.

The correspondence determining section 22 determines the correspondence between the character segments extracted from the character image and the dictionary segments stored in the dictionary 25. The representative point determining section 221 and the angle change detecting section 222 and a dynamic programming (DP) collation unit 223. The representative point determining unit 221 sets representative points to be used for DP collation at regular intervals on the character segment stored in the segment dividing unit 213 of the segment extracting unit 21, and the angle change detecting unit 222 sets three consecutive representative points. Then, the accumulated value of the angle change from the first representative point of the character segment to each representative point is calculated as the characteristic amount at each representative point.

FIG. 6 is a diagram for explaining detection of a change in angle. In FIG. 6, representative points (A, B, C,...) Are set at arbitrary intervals on the contour line (S), and a vector AB drawn from point A to point B and a vector AB drawn from point B to point C. Vector BC
And the angle between the vector AB and the vector BC (θ
2) is an angle change at the point B.

The angle change at the end point A of the outline (S), which is the initial value of the angle change, is defined as the angle (θ1) between the vector GA and the vector AB obtained by subtracting the point A from the center of gravity (G) of the character. , The characteristic amount at each representative point is represented by a value obtained by accumulating the angle change from the point A having the initial value of the angle change to each representative point,
For example, the feature value at point B is θ1 + θ2. The DP matching unit 223 compares the characteristic amount calculated by the angle change detection unit 222 between the representative point of the character segment selected by the representative point determination unit 221 and the representative point of the dictionary segment stored in the dictionary 25 and the dictionary 25. When the absolute value of the difference between the feature amount of the representative point of the dictionary segment and the feature amount stored in the dictionary segment is defined as the distance, one-dimensional DP matching is performed so that the closest representative point is associated.

FIG. 7 is a diagram for explaining the correspondence between segments. FIG. 7A shows a character segment, white circles represent representative points, and FIG. 7B a dictionary segment, black circles represent representative points, and arrows represent representative points on both segments. Indicates that they are compatible. The inspection point detection unit 23 is to calculate inspection information by finding inspection points on the character segments based on the correspondence determined by the correspondence determination unit 22. The inspection point determination unit 231 and the inspection information detection unit 232 It is composed of

In the inspection point determination unit 231, a reference point (d) set in advance on the dictionary segment shown in FIG.
The representative point on the character segment shown in FIG. 7C corresponding to (1, d2) is defined as the inspection point (c1, c2). In the DP matching, a many-to-many correspondence can be obtained. Therefore, when the number of the representative point on the character segment corresponding to the reference point on the dictionary segment is one, the representative point is set as the inspection point, and the dictionary segment is checked. When a plurality of representative points on the character segment correspond to the reference point above, the plurality of representative points are selected from among the plurality of representative points according to a preset condition such as a point close to an end point or a center point of the corresponding plurality of representative points. One point is an inspection point.

FIG. 8 is a flowchart showing a process of determining an inspection point. In FIG. 8, first, the value of the number (i) designating a specific reference point from a plurality of reference points is set to 1 (S80), and a reference point di is selected from the representative points on the dictionary segment (S81). Then, a representative point on the character segment corresponding to the reference point di is determined as the inspection point ci (S82).

Next, it is determined whether there is an unprocessed reference point (S83). If there is no unprocessed reference point, the inspection point determination process is terminated. Sets the number designating the reference point to i + 1 (S84), and thereafter returns to the process shown in S81 and repeats the subsequent processes of S82 and S83 in the same manner. In the inspection information detection unit 232, as the inspection information at the inspection point on the character segment, for example, 1
For one inspection point, the absolute position information of each inspection point indicating the position of the inspection point with respect to the entire character image, and for two inspection points, the distance between the inspection points For relative position information such as the direction and direction, and for two or more inspection points, information such as angle change and linearity between the inspection points is calculated.

The candidate deciding section 24 decides the category of the input character image based on the inspection information of the inspection point on the character segment obtained by the reference point inspecting section 23. 242. In the inspection information determination unit 241, the inspection point detection unit 23
The inspection information calculated by the inspection information detecting unit 232 is compared with the inspection information of the reference point on the dictionary segment stored in the dictionary 25 in advance.
It is determined whether the determination condition stored in advance in step 5 is satisfied or not satisfied.

The determination condition is, for example, as shown in FIG.
In the case where the angle change from the inspection point c1 on the character segment to c2 along the character segment is used as the inspection information, the character image of the character segment whose angle change is 60 degrees or more is assigned to the corresponding dictionary segment. Belongs to the category of the character candidate "2". The recognition result output unit 242 determines a category of the dictionary segment that satisfies the determination condition as a recognition result of the input image based on the determination of each dictionary segment, and outputs the category.

FIG. 9 is a flowchart showing the process of determining a recognition candidate. In FIG. 9, first, a value of a number (j) designating specific test information from a plurality of test information is set to 1 (S90), and an arbitrary reference point di on the dictionary segment is set.
The inspection information j is selected from the plurality of inspection information in (S91), and the inspection information j at the inspection point ci on the character segment corresponding to the reference point di is calculated (S92).

Next, the calculated inspection information j is compared with the determination condition of the inspection information j at the reference point di on the dictionary segment to determine whether or not the determination condition is satisfied (S93). Determines that the recognition result of the character segment is not the category of the dictionary segment (S9).
4) Subsequently, the next dictionary that has not been subjected to the recognition candidate determination processing is selected (S95), and the processing of the recognition candidate determination with the specific dictionary being processed is terminated.

If the determination condition is satisfied, it is determined whether there is unprocessed inspection information (S96). If there is no unprocessed inspection information, the recognition result of the character segment is determined to be the category of this dictionary segment. Judgment (S97),
The recognition result is output (S98), and the recognition candidate determination process ends. On the other hand, when there is unprocessed inspection information, the number designating the inspection information is set to j + 1 (S99), and thereafter, the process returns to S91, and the subsequent processes of S92 to S98 are similarly repeated.

The dictionary 25 stores a dictionary segment, a reference point, inspection information, and a judgment condition, which are information for specifying a category of an input character image, in a dictionary segment storage unit 251, a reference point storage unit 252, and an inspection point, respectively. Information storage unit 2
53, to store in the judgment condition storage unit 254 for each category. Note that the dictionary 10, the correspondence determining means 11, the inspection point detecting means 12, the candidate determining means 13 in FIG.
Respectively correspond to the dictionary 25, the correspondence determination unit 22, the inspection point detection unit 23, and the candidate determination unit 24 in FIG.

In the above configuration, a flow of processing operation of the character recognition device will be described. FIG. 3 is a flowchart showing a processing operation of the character recognition device according to the embodiment of the present invention. First, the form or the like is scanned by the scanner 201 of the character image input unit 20 and the input character image is binarized into a black and white binary image by the binarization unit 202 (S30). A process of extracting a character segment from the binary image (S31) is performed.

Then, from the plurality of dictionary segments stored in the dictionary segment storage unit, an arbitrary dictionary segment which has not been subjected to the correspondence relationship processing with the character segment is selected (S32). The correspondence process between the dictionary segments and the character segments is performed (S33). After the correspondence processing, the inspection point detection unit 23 determines an inspection point from the representative points taken on the character segment, and performs an inspection point determination process of calculating inspection information (S34).

Next, a comparison is made between the character segment and the dictionary segment based on the inspection information, and a candidate determination process is performed to determine whether the character segment matches a dictionary segment that is a recognition candidate (S35). It is determined whether or not the recognition has been performed (S36). If the candidate is determined, the recognition result is output and the recognition processing of the input character image is terminated. on the other hand,
If the dictionary segment selected in S32 is not determined as a candidate, it is determined whether there is an unprocessed dictionary that has not been subjected to the correspondence processing with the character segment (S37). S32 to select
And the subsequent steps S33 to S36 are repeated in the same manner.

If there is no unprocessed dictionary for which no correspondence processing has been performed, the input character image is determined to be unrecognizable, a recognition result indicating that recognition is not possible is output, and the recognition processing ends. .

[0033]

As described above, according to the present invention,
When matching a recognition target character image with a dictionary, the correspondence between the input character image and the input character image is determined for each dictionary to be compared, and an inspection point for comparing feature information is specified. There is an effect that a stable recognition rate can be obtained.

Further, a place to be collated for each category is set on the dictionary segment, and the characteristic information at the inspection point on the corresponding character segment to be recognized is calculated to determine the recognition. Can be greatly improved.

[Brief description of the drawings]

FIG. 1 is a principle block diagram of the present invention.

FIG. 2 is a block diagram of a character recognition device according to an embodiment of the present invention.

FIG. 3 is a flowchart illustrating a processing operation of the character recognition device according to the embodiment of the present invention.

FIG. 4 is a diagram illustrating extraction of a character segment.

FIG. 5 is a diagram illustrating detection of an end point.

FIG. 6 is a diagram illustrating detection of an angle change.

FIG. 7 is a diagram illustrating the correspondence between segments.

FIG. 8 is a flowchart showing a process of determining an inspection point.

FIG. 9 is a flowchart illustrating a process of determining a recognition candidate.

FIG. 10 is a block diagram of a conventional character recognition device.

FIG. 11 is a diagram illustrating character recognition based on a geometric shape of a character image.

[Explanation of symbols]

 DESCRIPTION OF SYMBOLS 10 ... Dictionary 11 ... Correspondence determination means 12 ... Inspection point detection means 13 ... Candidate determination means

──────────────────────────────────────────────────続 き Continuation of the front page (72) Inventor Haruo Akimoto 1015 Kamikodanaka, Nakahara-ku, Kawasaki City, Kanagawa Prefecture Inside Fujitsu Limited (56) References JP-A-58-66176 (JP, A) JP-A-3-282896 (JP, A) JP-A-61-190682 (JP, A) JP-A-61-198381 (JP, A) JP-A-2-217981 (JP, A) JP-A-58-163080 (JP, A) IEICE Transactions, Vol. J67-D No. 3 p. 351-358 (1984) Hirozo Yamada "Contour DP Matching Method and Application to Handwritten Kanji Recognition" (58) Fields investigated (Int. Cl. ⁷ , DB name) G06K 9/46-9/62 JICST file (JOIS)

Claims (1)

(57) [Claims] 1. From a character contour extracted from a character image
A dictionary that stores a line segment obtained by cutting off the end as a dictionary segment, and stores the characteristics of the dictionary segment at reference points taken on the dictionary segment as dictionary characteristic information of the character image; Character outline extracted from the character image of the character
The line segment obtained by cutting the end from the line
A specific representative point on the character segment
Using the cumulative value of the angle change from the point to each representative point
The legal, and the representative point on the character segment, the correspondence determining means to take correspondence between the representative point on the dictionary segment, based on the correspondence relationship obtained by the relationship determining means,
A point on the character segment corresponding to the reference point on the dictionary segment is set as an inspection point, and the characteristic information of the character segment at the inspection point corresponding to the dictionary characteristic information of the dictionary segment at the reference point An inspection point detection unit to be extracted; and a recognition candidate for the character image to be recognized by comparing the feature information of the character segment extracted by the inspection point detection unit with dictionary feature information of a corresponding reference point on the dictionary segment. And a candidate deciding means for deciding a character.