JPH0731711B2 - Optical character reader - Google Patents

Description

Description: TECHNICAL FIELD OF THE INVENTION The present invention relates to an optical character reading device for performing reading processing of free handwritten characters.

[Technical background of the invention and its problems]

In recent years, optical character readers (hereinafter referred to as OCR) have
There is a system that can read handwritten characters. In the OCR of such a system, a large number of standard patterns are prepared in advance in the dictionary memory to deal with the deformation of characters due to individual habits.

However, in the OCR of the above-mentioned method, when reading freehand written characters written by unspecified number of people,
Since the number of standard patterns to prepare is huge, a huge amount of dictionary memory is required. If the storage amount of the standard pattern is reduced, the OCR reading accuracy will be degraded. Conventionally, there has been a method in which post-processing such as text processing is performed to improve the reading accuracy with respect to the deterioration of the reading accuracy. However, in this case, there is an inconvenience that cannot be applied when the handwritten characters are pure numbers only.

[Object of the Invention]

The present invention has been made in view of the above circumstances, and an optical character reading device capable of performing a reading process with high accuracy in a reading process of freehand-written characters without using a dictionary memory having a particularly large capacity. The purpose is to provide.

[Outline of Invention]

The present invention relates to a feature extraction unit that extracts a feature of a character pattern for each character that is photoelectrically converted by scanning a sheet, and a feature memory that stores at least one sheet of the feature of a character pattern output from the feature extraction unit. , A classifying unit that obtains the similarity between the characteristics of the character patterns read from the characteristic memory, and classifies the characteristics of the character patterns having a certain value of similarity as a group of the characteristics of the character pattern corresponding to a specific character. An average pattern extractor for extracting an average pattern feature for each selected group, and an optical disc having an identifying unit for identifying a specific character by comparing the extracted average pattern feature with a standard pattern stored in advance in a dictionary memory. It is a static character reader. As a result, the free handwritten character reading process can be performed with high accuracy without using a particularly large-capacity dictionary memory.

Example of Invention

An embodiment of the present invention will be described below with reference to the drawings. In the figure, reference numeral 1 denotes a photoelectric conversion unit which converts a scanning pattern A of free handwritten characters obtained by scanning a sheet into a binarized signal, and also performs an appropriate preprocessing. The feature extraction unit 2 extracts the features of the character pattern for each character sent from the photoelectric conversion unit 1. The feature memory 3 stores the features extracted by the feature extraction unit 2 for each character, and has a storage capacity of, for example, one sheet. The classification unit 4 classifies the character pattern features read from the feature memory 3 into a plurality of groups for each character. 5
Is a judgment processing unit, which judges the end of the classification process in the classification unit 4, and when it is judged that the classification process is completed, the character pattern feature group for each group is sent from the classification unit 4 to the average pattern extraction unit 6. If the determination processing unit 5 determines that the classification process is impossible, that is, the similarity of a certain value or more is not obtained based on the similarity calculation result, and the character pattern features cannot be classified into a specific group, the determination unit 5 The control signal B is output to 7.
Based on the control signal B of the determination processing unit 5, the determination unit 7 is a control signal that causes the process from the photoelectric conversion unit 1 to be performed again.
A control signal C2 that causes the processing from C1 or the feature extraction unit 2 to be performed again is output. The average pattern extraction unit 6 extracts average pattern features of each character group. 8
Is an identification unit that identifies and outputs a character from the pattern feature from the average pattern extraction unit 6 based on a standard pattern stored in advance in the dictionary memory 9.

The operation of such a configuration will be described. Now, suppose that one sheet of freehand handwriting is line-by-line for photoelectric conversion unit 1.
Is converted into a binary signal and sent to the feature extraction unit 2. The feature extraction unit 2 extracts the features of the character pattern cut out for each character. The feature memory 3 stores the character pattern features for one sheet or a plurality of sheets (however, it is known that the same individual has written a character) for each character. Further, the classification unit 4 reads the character pattern features from the feature memory 3 and classifies the character pattern features into groups of similar characters. In this case, the classification unit 4 judges and classifies the similarity based on the matching between the character pattern features or the degree of difference between the feature codes. The determination processing unit 5 determines that the classification process is completed when the similarity calculation performed by the classification unit 4 yields a similarity equal to or greater than a certain value. Further, when it is determined that the classification processing is impossible because the similarity of a certain value or more is not obtained, the control signal B corresponding to the content is output. For example, when the photoelectric conversion unit 1 does not sufficiently reproduce the character image by signalization, the determination processing unit 5 outputs the control signal B instructing the process from the photoelectric conversion unit 1 to the determination unit 7 again. The determination unit 7 outputs the control signals C1 and C2 so that the process from the photoelectric conversion unit 1 or the process from the feature extraction unit 2 is performed again based on the control signal B.

By the way, when the determination processing unit 5 determines that the classification processing by the classification unit 4 is completed as described above, the classification unit 4 supplies the character pattern feature group for each group to the average pattern extraction unit 6. The average pattern extraction unit 6 calculates and obtains an average pattern feature of the character pattern feature group. In this case, the average pattern extraction unit 6 obtains the average pattern feature by a method such as calculating the center coordinates of the distribution when the character pattern feature group is expressed in the metric space. The identification unit 8 identifies which character the average pattern feature obtained by the average pattern extraction unit 6 corresponds to, based on a standard pattern stored in advance in the dictionary memory 9.

In this way, the free handwritten characters recorded on the paper can be read. In this case, in the present invention,
The classification unit 4 classifies character pattern features for at least one sheet into groups for each character by similarity calculation. Therefore, the classified groups are divided into different categories, and the same answer cannot be obtained when the identifying unit 8 identifies the character from the average pattern feature of each group. Therefore, free handwritten characters can be reliably read.

〔The invention's effect〕

As described above in detail, according to the present invention, in the free handwritten character reading process, the character pattern features for at least one sheet are classified into groups for each character, and the average pattern features are obtained for each group. Therefore, the characters can be surely identified. Therefore, as a result, the free handwritten character reading process can be performed with high accuracy without using a large standard pattern stored in the dictionary memory having a particularly large capacity.

BRIEF DESCRIPTION OF THE DRAWINGS FIG. 1 is a block diagram showing the configuration of an optical character reading device according to an embodiment of the present invention. 2 ... feature extraction unit, 3 ... feature memory, 4 ... classification unit, 6 ... average pattern extraction unit, 7 ... dictionary memory, 8 ... identification unit.

Claims (1)

[Claims] 1. A feature extraction unit for extracting a feature of a character pattern for each character that is photoelectrically converted by scanning a sheet, and stores the feature of the character pattern output from the feature extraction unit for at least one sheet. And a similarity between the features of the character pattern read from the feature memory, and the features of the character pattern having a similarity of a certain value or more as a group of features of the character pattern corresponding to a specific character. A classifying unit for classifying, an average pattern extracting unit for extracting an average pattern feature for each of the groups classified by the classifying unit, the average pattern feature extracted by the average pattern extracting unit, and a pre-stored dictionary memory. An optical character reading device comprising: an identification unit that matches the standard pattern and identifies the specific character.