JPS62120586A - Character recognizing device - Google Patents

Abstract

PURPOSE:To recognize a laterally written character such as alphabets included in a vertical written document by recognizing a rejected character through the use of a dictionary for laterally written characters after its picture is rotated by 90 deg.. CONSTITUTION:Picture data from a scanner 2 is accumulated in a buffer 4, and accessed by a vertical and lateral direction decision part 6 and a line segment part 8. If an inputted sentence is decided to be vertical written, it is inputted to a feature extraction part 14 without being rotated 90 deg.. The feature vector of the sentence is inputted to a matching part 16, and its input and the feature vectors of a common dictionary D1 and a laterally written character dictionary D3 are calculated. If a rejected character occurs because a character with a minimum distance cannot be found, a rotating part 12 rotates the input character by 90 deg.. The picture data of the input character is transmitted to the feature extraction part 14, and its feature vector is calculated in the matching part 16 based on the common dictionary D1 and the laterally written character dictionary D2. Finally the code of the character with a minimum distance is outputted as a recognized result.

Description

DETAILED DESCRIPTION OF THE INVENTION [Technical Field] The present invention relates to a character recognition device, and more particularly to a character recognition device capable of recognizing characters in vertically written documents.

[Prior art]

Character recognition devices are generally made for horizontally written documents, but
Some character recognition devices for vertically written documents have also been developed.

However, in conventional character recognition devices for vertically written documents,
When a vertically written document contains horizontally written characters.

It was impossible to recognize the horizontally written characters.

For example, a vertically written document like the one shown in FIG. 3 is not uncommon, but the characters in the Roman character string on the second line of this document could not be recognized.

〔the purpose〕

An object of the present invention is to provide a character recognition device that can correctly recognize horizontally written characters such as European characters included in a vertically written document.

〔composition〕

The character recognition device of the present invention, which has been made to achieve this object, is equipped with a dictionary for vertically written characters, a dictionary for horizontally written characters, and a means for rotating a character image by 90 degrees. When a reject character is generated during character recognition of a written document, the image of the reject character is rotated by 90 degrees, and then recognition is attempted using a dictionary of horizontally written characters.

〔Example〕

An embodiment of the present invention will be described below with reference to the drawings.

FIG. 1 is a schematic block diagram showing an embodiment of the character recognition device of the present invention. In this figure, numeral 2 denotes a scanner that separates and reads an input document into pixels, and outputs black and white binary image data. This image data is temporarily stored in the buffer 4. This buffer 4 is accessed by an aspect determination section 6 and a line cutting section 8.

The vertical/horizontal determining unit 6 determines whether the input document is a vertically written document or a horizontally written document. This type of vertical/horizontal determination can be performed using various methods.

In this embodiment, the orientation is determined by the following method.

When examining the run characteristics of a document image, it is found that short vertical white runs occur more frequently in vertically written characters, while short horizontal white runs occur more frequently in horizontally written characters. Therefore, the aspect determination unit 6 accesses the buffer 4 to read the image data of the input document, measures the frequency of occurrence of short white runs in the vertical direction and short white runs in the horizontal direction, and uses the measurement results to determine whether the document is vertically written or horizontally written. Determine. The details of such vertical/horizontal discrimination are described in Japanese Patent Application Laid-Open No. 56-149674 entitled "Identification Method of Image Characteristics".

The line cutting unit 8 accesses the buffer 4 to read image data, cuts out lines of the input document, and inputs the image data of each line to the character cutting unit 10. This line cutting process is performed, for example, by a known projection method.

Since this line cutting process requires changing the line cutting direction for vertical and horizontal writing, the vertical/width determining unit 6 notifies the line cutting unit 8 of the results of the vertical/width determination.

The character cutting section 1o cuts out image data of individual characters from the line-by-line image data inputted from the line cutting section 8, and the character cutting process is performed, for example, by a known projection method. The extracted character image data is transferred to the 90 degree rotation section 1.
2 to the feature extraction unit 14.

This 90 degree rotation unit 12 normally transmits the character image data as is (without performing a 90 degree rotation operation) to the feature extraction unit 14, but when 90 degree rotation is instructed by the matching unit 16, , provides image data obtained by rotating the character image by 90 degrees to the feature extraction unit 14.

FIG. 2 shows the configuration of the 90 degree rotating section 12. As shown in this figure, the 90 degree rotation unit 12 has character image memories 3o and X that can store image data for one character.

The main elements are Y address counters 31 and 32. When writing image data to the character image memory 30, the X address counter is sequentially incremented from zero and when it reaches the maximum value, the X address counter 31 is incremented sequentially from zero.
When is cleared to zero, the Y address counter is incremented by 1. Such address updating is repeated until the X address counter 32 reaches the maximum value.

Reading of an image from the character image memory 30 is normally performed while incrementing the X and Y address counters 31 and 32, as in the case of writing.

On the other hand, when a 90 degree rotation is instructed, the X address counter 32 is decremented from the maximum value,
By updating the address by incrementing the X address counter 31 by 1 from zero when it reaches zero, the feature extraction unit 14 extracts the image data of the character rotated by 90 degrees.
send to

The feature extraction unit 14 extracts the features of the input character image data to create a feature vector, and the matching unit 1
Send to 6. The matching unit 14 performs a matching operation between the feature parameters and the feature vectors of each character registered in the dictionary 18, and searches for a character with the minimum distance.

The dictionary IF 18 is basically composed of a dictionary for vertically written characters and a dictionary for characters, but in this embodiment, in order to reduce the dictionary capacity, feature vectors are common to both vertically and horizontally written characters. The structure is such that the dictionaries for characters are integrated into one common dictionary D1, and the dictionaries for other characters are separated into a dictionary for horizontally written characters D2 and a dictionary for vertically written characters D3.

Since the dictionaries and the like used are different between vertically written documents and horizontally written documents, the matching section 16 is notified of the result of the vertical/horizontal determination from the vertical/width determining section 6.

Next, the overall operation will be explained. First, an explanation will be given of the operation when the aspect determination unit 6 determines that the input document is a horizontally written document.

In this case, line cutting and character cutting are performed as a horizontally written document, and the image data of the cut out characters is directly input to the feature extraction unit 14 through the 90 degree rotation unit 12, and the feature vectors of the characters are input to the matching unit 16. sent to.

In this case, the matching section 16 operates in horizontal writing mode,
Calculates the distance between the feature vector of the input character and the feature vectors of the common dictionary D1 and the dictionary for horizontal characters D2, searches for a character with the minimum distance, and if the distance is less than a predetermined value, outputs the code of that character, Otherwise, output a reject code.

Next, the operation when the input document is determined to be a vertically written document will be explained. In this case, line cutting and character cutting are performed as a vertically written document, and the matching unit 16 operates in vertical writing mode.

The extracted character image data is input to the feature extraction section 14 without being rotated by 90 degrees, and its feature vector is input to the matching section 16.

The matching unit 16 performs a matching operation between the feature vector of the input character and the feature vectors of the common dictionary D1 and the dictionary for vertical characters D3, and searches for a character with a minimum distance less than or equal to a predetermined value. If such a character is found, the matching unit 16 outputs the character code and proceeds to recognize the next character.

But if no such character is found.

In other words, if a reject character occurs, matching section 1
6. A try instruction is sent. In response to this instruction, the 90 degree rotation section 12 performs a 90 degree rotation operation on the input character that has become a rejected character, and sends the image data of the input character after the 90 degree rotation to the feature extraction section 14 . Feature extraction unit 14
extracts the features of the 90 degree rotated character and sends the feature vectors to the matching unit 16. The matching section 16 is
Next, a matching operation is performed on the feature vector with the common dictionary D1 and the dictionary for horizontal characters D2. If a character with a minimum distance less than a predetermined value is found through this second matching operation, that character code is used as the recognition result. output and proceed to recognizing the next character. If a character with a minimum distance equal to or less than a predetermined value is not found this time, the matching unit 16 finally determines the input character as a reject character, outputs a reject code, and proceeds to recognize the next character.

In this way, character recognition of vertically written documents is performed using the vertical writing dictionary (in this embodiment, the vertical writing dictionary D3 and the common dictionary WDL), but if a reject character is generated in such recognition, , for characters obtained by rotating the input character 90 degrees as reject characters, the horizontal writing character dictionary (in this embodiment, the horizontal writing character dictionary D2 and the common dictionary Di
) character recognition is attempted.

For example, the Roman portion of the second line of the vertically written document shown in FIG. 3 cannot be recognized in the first character recognition operation using the dictionary for vertically written characters, and a reject character is generated. However, it is clear that each character in such a Roman part can be correctly recognized by a second character recognition operation in which the input character is rotated by 90 degrees and a dictionary for horizontally written characters is used.

〔effect〕

As explained in detail above, the present invention includes a dictionary of vertically written characters, a dictionary of horizontally written characters, and a means for rotating a character image by 90 degrees, and uses the dictionary of vertically written characters to rotate characters of a vertically written document. If a reject character occurs during recognition, the image of the rejected character is rotated 90 degrees and then recognition is attempted using a dictionary of horizontally written characters. A character recognition device that can also recognize characters can be realized.

[Brief explanation of drawings]

Fig. 1 is a schematic block diagram showing an embodiment of the character recognition device of the present invention, Fig. 2 is a schematic block diagram of a 90 degree rotation section, and Fig. 3 is an example of a vertically written document containing a mixture of horizontally written character strings. It is a diagram. 2... Scanner, 4... Buffer, 6... Vertical/horizontal determination section, 8... Line cutting section, 10... Character cutting section, 12... 90 degree rotation, 14... feature extraction section,
16...Matching section, 18...Dictionary, Dl.
・・Common dictionary, D2・・Dictionary for horizontal writing characters, D3・
・Dictionary for vertical writing characters. 1st ward I Otsu 2nd figure? -F

Claims (1)

[Claims] (1) When a reject character occurs during character recognition of a vertically written document using a dictionary of vertically written characters, a dictionary of horizontally written characters, and a means for rotating the character image by 90 degrees. , a character recognition device characterized in that after rotating an image of the rejected character by 90 degrees, recognition is attempted using a dictionary of horizontally written characters.