JP2851865B2 - Character recognition device - Google Patents

Description

Description: TECHNICAL FIELD The present invention relates to a character recognition device that converts a character image into a character code.

[Conventional technology]

2. Description of the Related Art Conventionally, a character recognition device calculates a distance between a feature pattern extracted from a character image and a standard feature pattern whose character code is known, and outputs a character code having a similarity between the two feature patterns as a recognition result of the character image. Document HA is one of the character recognition methods used in character recognition devices.
Glucksman “Classification of mixed font alphabetic
s by characteristic loci "Digestol 1st Ann.IEEE Com
puter Conf., pp. 137-141, 1967, the character recognition method (hereinafter referred to as Graxman's method) captures the topological features of characters well, so it uses simple recognition processing for alphanumeric characters and characters. It is known that character recognition of about katakana can be performed relatively successfully.

In the method of Glaxman, focusing on the background portion of the character image, a half line is extended in the vertical and horizontal directions from the white image forming the background portion, and the number of times of intersection with the character line is counted. Then, the count value is divided into three stages of 0 times, 1 time, 2 times or more, and coded into numerical values of “0”, “1”, and “2”, respectively, and this code is coded for each direction for one white pixel. You. Next, a combination of the above codes (surrounding state information) for 3 ⁴ = 81 character lines for each white pixel is created in advance as a table, and the surrounding information code is extracted from the character image to be recognized. The number of white pixels having the same surrounding information code as the surrounding information code is given as an 81-dimensional feature vector. Next, character recognition is performed by comparing this feature vector with a standard feature vector whose character code is known.

[Problems to be solved by the invention]

However, when character recognition is performed by performing a distance calculation between a feature pattern (vector) extracted from a character to be recognized and a standard feature pattern extracted from a character whose character code is known, without being limited to the above-mentioned Graxman method, Characters with similar shapes may be erroneously recognized. For example, in the above-mentioned Graxman's method, set J to n and 9 to? In some cases, he mistaken for Shi, N for Shi, and Mi for 8.

Accordingly, it is an object of the present invention to provide a character recognition device capable of solving the above-mentioned problems and improving the character recognition accuracy without significantly changing the configuration of the conventional character recognition device. .

[Means for solving the problem]

In order to achieve such an object, the first embodiment of the present invention calculates a distance between a characteristic pattern extracted from a character image to be recognized and a standard characteristic pattern prepared for each character code, and calculates a plurality of distances approximate to the characteristic pattern. Character identification means for outputting the character codes of the standard patterns as character recognition candidates, and at least one of the length and the number of contour lines constituting the character image by using an image recognition method from the character image to be recognized. Detecting means for detecting one of the contour feature information; storage means for storing standard contour feature information preset for each character code for collating the contour feature information; The standard contour feature information corresponding to the character code is extracted from the storage means, and the extracted standard contour feature information matches the contour feature information detected by the detection means. Characterized in that comprises a selecting means for selectively outputting the character code that.

Further, the second mode of the present invention is characterized in that the number included in the second contour information has a fluctuation upper limit value and a lower limit value.

(Operation)

In the present invention, attention is paid to the fact that the length or the number of contour lines constituting the character is determined for each character, and a plurality of recognition candidate characters are calculated by calculating a distance from a feature pattern extracted from each of a recognition target character and a standard character. Based on selection and outline feature information, character recognition results are selected and collated from the selected candidate characters.For example, when performing character recognition using the Graxman method, more intuitive human observation Close recognition can be achieved and high-precision recognition is possible.

〔Example〕

 Hereinafter, the present invention will be described in detail with reference to the drawings.

FIG. 1 shows a circuit configuration of a first embodiment of the present invention. In FIG. 1, reference numeral 1 denotes a reading unit using a photoelectric conversion element for reading a document and converting it into binary image data. Reference numeral 2 denotes an image memory for storing binary image data. 3 is an image memory 2
A character extraction circuit using a central processing unit (CPU) for extracting the binary image data stored in the CPU into a size of one character, and 4 extracts a character feature vector from the extracted character image data by the Graxman method. This is a feature extraction circuit that extracts and sends the extracted feature vector to the first identification circuit 5, and uses a CPU.

A CP 5 receives the feature vector and calculates a distance between the feature vector and an average vector of each character in the dictionary unit 6.
This is a first identification circuit using U, and selects, for example, three candidate characters from the smallest distance, and sends the character codes of the three candidate characters to the matching circuit 8.

Reference numeral 7 denotes a contour line number counting circuit using a CPU for extracting contour lines from the cut-out character image data in the image memory 2 and calculating coefficients of the number of inner contours and the number of outer contours. Reference numeral 8 denotes a matching circuit using a CPU, which compares the number of inner contour lines and the number of outer contour lines obtained by the contour line counting circuit 7 with the number of inner contour lines and the number of outer contour lines for each character in the contour line information table. And
The best matching character among the three candidate characters obtained by the first recognition circuit 5 is selected and its character code is output. An output buffer 10 stores a character code.

 FIG. 2 shows the contents of the contour line information table.

In FIG. 2, the contour line information table 9 stores the character code of the character to be recognized and the upper limit of the number of inner contour lines and the upper limit of the change of the number of outer contour lines corresponding to the character code. Here, the upper limit of variation means the upper limit of the number of contour lines that can be statistically due to the variation of handwritten characters. For example, the letter “E” has the number of outer contour lines “1” in normal writing
However, if the characters are written apart from each other, the number of outer contours will be "2".
Therefore, the upper limit of the fluctuation is “2”. If the lower limit of variation is used in addition to the upper limit of variation, the accuracy of character recognition is increased.

Returning to FIG. 1, the circuit operation of the embodiment of the present invention will be described with reference to the flowcharts of FIGS. 3 (1) and 3 (2).

When the reading unit 1 inputs a document image from a document, the document information is converted into a binary image and stored in the image memory 2 (step S1 in FIG. 3). Next, the character cutout circuit 3 operates, and the binary image data for each document is cut out into character image data for each character and written in the margin of the image memory 2 (step S2).

Next, the feature extraction circuit 4 operates to create a feature vector from the character image data by the method of Graxman, and outputs this feature vector to the identification circuit 5 (step S11). Next, the identification circuit 5 operates, calculates the Euclidean distance between the average vector and the feature vector of each character in the dictionary unit 6, and searches (sorts) three candidate character codes in ascending order of distance value (steps S12 to S16). ). The operation so far is the same as that of the conventional character recognition circuit. Next, the three candidate character codes are passed to the matching circuit 8 in the order of the smallest distance value.

On the other hand, at the same time as the operation of the feature extraction circuit 4 starts, the number of contour lines counting circuit 7 starts to calculate the number of inner contour lines and the number of outer contours of the character image data by a well-known contour tracking method. It is sent to the matching circuit 8 (step S20). When the input of information from the contour line number counting circuit 7 and the identification circuit 5 is completed, the matching circuit 8 performs the next matching process. here,
Entered the inner contour number n _i, the outer contour number n _o input, the first candidate character, the second option character and the third candidate characters as the corresponding inner contour number and upper outer contour maximum number of contour information table 9 _{Are denoted as} N _1i , N _1o , N _2i , N _2o , and N _3i , N _3o , respectively. That is, the matching process (1) The above numerical values are compared, and the comparison result is n _i ≦ N _1i and n _1o ≦
In the case of _N1o , the first candidate character is output to the output buffer 10 as the final recognition result as it is as the final recognition result (step S22 → S26).

When the comparison result is n _i > N _1i or n _o > N _1o , the following collation processes (2) and (3) are performed.

(2) If the magnitude relation of the number of contours is n _i ≦ N _2i and n _o ≦ N _2o , the first candidate character and the second candidate character are exchanged. Otherwise, the next matching process (3) is performed (step S23 → S
twenty five).

(3) If the relationship of the number of contours is n _i ≦ N _3i and n _o ≦ N _3o , the first candidate character and the third candidate character are switched (step S24 → S25).

After performing the above-described matching processes (1) to (3), the matching circuit 8 outputs the character code of the first candidate character to the output buffer 10 (Step S26).

In the present embodiment, the number of recognition candidates is limited to three. However, the number may be set to a predetermined number as desired. All the character codes having the distance calculation result may be set as recognition candidates.

[Other embodiments]

Next, a second embodiment will be described. This example is an example in which the inner contour length and the outer contour length are used as the contour information instead of the number of contours.

In this case, as shown in FIG.
Replaces the contour length calculation circuit 7 ', and the content of the contour line information table changes from the contour line to the contour line length.

In this example, as in the first embodiment, the identification circuit 5 calculates the distance between the average vector for each character and the input feature vector, and sends three character codes having an average vector with a small distance and the distance value to the matching circuit 8. Sent. These distance values are represented by d ₁ , d ₂ and d ₃ in order. The order indicated by c _1, c _2, c ₃ a character code selected as a recognition candidate. Inner contour line length of the calculated input character image in the contour line length calculating circuit 7 and the outer contour line length respectively indicated by l _i, l _o.

In the matching circuit 8, the distance with respect to the contour length for each of the three candidate characters is obtained by the following equation.

e _k = (l _i −L _i ^(k) ) ² + (l _o −L _o ^(k) ) ² where e _k is the distance of the candidate character C _{k to be obtained} , L _i ^(k) , L _o ^(k) is the average length of the average length and the outer contour of the inner contour line of the contour line information table for the candidate character C _k (see FIG. 5). Then, a new distance _Dk is calculated by the following equation using the predetermined weight ω.

D _k = d _k + ωe _k ( _k = 1,2,3) Then, the corresponding character code C _k of the minimum value among D _k ( _k = 1,2,3) is sent to the output buffer.

〔The invention's effect〕

As described above, according to the present invention, attention is paid to the fact that the length or the number of contour lines constituting the character is determined for each character, and the distance calculation with the feature pattern extracted from each of the recognition target character and the standard character is performed. Is selected based on the outline information, and the character recognition result is selectively collated from among the selected candidate characters.For example, when performing character recognition using the method of Graxman, A classification close to intuitive human observation can be achieved, and highly accurate recognition is possible.

Further, by setting a lower limit value in addition to the upper limit value to the number of contour lines, character recognition can be performed accurately even when the contour lines forming the character are separated from each other.

[Brief description of the drawings]

FIG. 1 is a block diagram showing the circuit configuration of the first embodiment of the present invention, FIG. 2 is an explanatory diagram showing the contents of the contour line information table 9 of the first embodiment of the present invention, and FIGS. ) Is a flowchart showing the operation procedure of the first embodiment of the present invention, FIG. 4 is a block diagram showing the circuit configuration of the second embodiment of the present invention, and FIG. 5 is the contents of the contour line information table of the second embodiment of the present invention. FIG. 1 ... Reading unit, 2 ... Image memory, 3 ... Character extraction circuit, 4 ... Feature extraction circuit, 5 ... Identification circuit, 6 ... Dictionary unit, 7 ... Contour line counting circuit, 8 ... Collation Circuit 9 ... Contour line information table 10 ... Output buffer.

Continued on the front page (58) Fields surveyed (Int.Cl. ⁶ , DB name) G06K 9/46 G06K 9/62 G06K 9/68 G06K 9/03

Claims (2)

(57) [Claims] A distance calculation is performed between a feature pattern extracted from a character image to be recognized and a standard feature pattern prepared for each character code, and the character codes of the plurality of standard patterns approximate to the feature pattern are converted into a character code. Character identification means for outputting as a recognition candidate, and detecting, from the character image to be recognized, at least one of contour length feature information of the length and the number of contours constituting the character image by using an image recognition technique. Detecting means; storage means for storing standard contour feature information preset for each character code for collating the contour feature information; and the plurality of character codes corresponding to the plurality of character codes output from the character identifying means. Standard outline feature information is extracted from the storage unit, and the extracted standard outline feature information and the outline feature information detected by the detection unit are extracted. Matching character recognition apparatus, characterized in that it comprises a selection means for selectively outputting the character code. 2. The character recognition device according to claim 1, wherein the number of contour lines included in the standard contour line feature information has a fluctuation upper limit value and a lower limit value.