JPH04182788A - Method and device for recognizing character - Google Patents

Abstract

PURPOSE:To exactly execute the segmentation of characters and to reduce memory capacity for a dictionary by executing the shift processing of data. CONSTITUTION:When an operator indicates it from a character type selecting means 110 that a character type is inclined or a CPU 1 or an identification calculation part 6 judges the character type is inclined, the shift processing is executed for the data to be transferred from a temporary buffer 109 to an image memory 103. The shift processing of the data is executed by transferring the contents of the low-order byte of an i-th byte in the temporary buffer to the low-order byte of the i-th byte in the (j+iX2)th column of an image buffer 202 and transferring the contents of the high-order byte to the high-order byte of the i-th byte in the (j+iX2+1)th column in the i-th transfer from the temporary buffer 109 to the image memory 103. Thus, the data for one character can be exactly segmented, and the memory capacity for the recognition dictionary is reduced.

Description

DETAILED DESCRIPTION OF THE INVENTION [Field of Industrial Application] The present invention relates to a character recognition method and apparatus that can also recognize characters with slanted fonts.

BACKGROUND ART A conventional character recognition device is generally configured as shown in FIG. 401 is a scanner that converts character images into analog electrical signals; 402 is a scanner that converts analog signals from the scanner into 2
403 is an image memory for storing the binarized image data; 404 is a character cutting means for extracting a character area for each character from the character string on the image memory; 405 is a Feature extraction means 407 extracts character-specific features according to a predetermined algorithm.
406 is a recognition dictionary that stores statistics such as average value and variance obtained as a result of performing the same feature extraction process on the learning data as in 05; A collation means compares the content and selects the closest candidate character type as a recognition result, and 408 is an output means for transmitting the code of the candidate character type as a recognition result to a display or an external device.

[Problem to be solved by the invention] Figure 3 shows an example of Roman characters, but the characters include not only fonts with no slope (for example, Roman fonts) like 302, but also
Slanted fonts like 301 (e.g. italics)
However, conventional character recognition devices did not have the technology to handle slanted fonts like 301. Therefore, the following problems arise.

(1) Adjacent characters may overlap, and adjacent characters may intrude into the cutout data for one character, causing misrecognition.

(2) It is necessary to have multiple recognition dictionaries according to the fonts, and the memory capacity for the recognition dictionaries is extremely large.

(3) Typical feature extraction algorithms for character recognition focus on capturing the local slope of character lines.

Therefore, even if you try to create a recognition dictionary for slanted fonts, the algorithm optimized for normal, non-slanted characters with many horizontal and vertical lines will not be able to adequately recognize the characteristics of characters with many slanted lines. They are often not caught.

[Means for Solving the Problems] In order to solve the above problems, according to the present invention, an instruction is given to shift image information in a desired manner, the image information is stored one column at a time, and the image information is A character recognition method is provided, characterized in that, when a shift is instructed, the stored image information is controlled to be shifted and transferred.

In order to solve the above problems, according to the present invention, there is provided an instruction means for instructing to shift the image information in a desired manner, a storage means for storing the image information one column at a time, and a storage means for storing the image information one column at a time. A character recognition device is provided, characterized in that it has a transfer control means that controls to shift and transfer the image information stored in the storage means when an instruction is given to shift the information.

[Example] FIG. 1(A) is a block diagram showing an example of the present invention, and 101
to 108 are the same as 401 to 408, respectively, of the conventional example in FIG. In this embodiment, 102 binarization means and 10
A temporary buffer 109 for storing one column of image data read by a scanner is provided between the three image buffers, and a font selection means 110 is provided for selecting a font.

FIG. 1(B) is a block diagram showing the configuration of this embodiment, and the correspondence with the configuration diagram of FIG. 1(A) will be explained here.

1 is a CPU, which controls a binarization means 102 that binarizes image information read from the scanner 8 and other processing. 2 is a keyboard (KB), 3 is a pointing device (P, D,), and font selection means 1 where an operator specifies the font by looking at the font of the character to be recognized.
10, and is also used to give instructions for other tasks. 4 is read-only memory (ROM)
), and a recognition dictionary 107 used when recognizing characters is stored in advance (.5 is a memory, and the scanner 101
It plays the role of a temporary buffer 109 and an image memory 103 that temporarily stores image information read from the image data column by column.

Reference numeral 6 denotes an identification calculation unit that performs calculations for the main part of character recognition, character extraction means 104 that extracts characters from data in the image memory 103, and feature extraction means 105 that extracts features of the characters from the extracted data. , collation means 1 for collating the extracted feature data with the recognition dictionary 107
06 processing is performed. Reference numeral 7 denotes a CRT, which is an output means 108 for outputting the recognition results, and is a display means for displaying the progress before the recognition results are output, data prompting instructions to the operator, and the like. 8 is a scanner (SCAN), and the first
This corresponds to the scanner 101 in Figure (A). 9 is an interface with the scanner 8, 5CAN 1. /F.

FIG. 2 shows a temporary buffer 109 and an image memory 10.
A method of transferring data from the temporary buffer 109 to the image memory 103 will be explained using an example of the data of No. 3.

First, image data inputted by the scanner 101 is binarized by the binarization means 102. In this binarization, one pixel on the image data corresponds to one bit, and a black pixel is converted to a bit of 1, and a white pixel is converted to a bit of 0.

In FIG. 2, 201 is a temporary buffer, 20
2 represents the state of data in the image memory. The data on the temporary buffer 201 is called the 0th byte, 1st byte, . . . , i-th byte, . . . in order from the top.
Among the bytes, the upper side is the lower byte and the lower side is the upper byte.

First, input from the scanner 101 is input to the binarization means 102.
The binarized image data is sequentially sent to the temporary buffer 109.

When the operator does not give an instruction that the font is slanted from the font selection means 110, or when the CPUI or identification calculation unit 6 determines that the font is not slanted, one column of data is stored in the temporary buffer 109. Data is sequentially transferred to the first column of the image buffer 103 each time the data is stored.

However, when the operator gives an instruction that the font is slanted from the font selection means 110, or when the CPUI or identification calculation unit 6 determines that the font is slanted, the data is transferred from the temporary buffer 109 to the image memory 103. Performs shifting processing of data to be transferred.

In the data shifting process, in the i-th transfer from the temporary buffer 109 to the image memory 103, the contents of the lower byte of the i-th byte of the temporary buffer are transferred to the lower byte of the i-th byte of the (j+1x2)th column of the image buffer 202. The content of the upper byte is the (j+ix2+1
) is transferred to the upper byte of the i-th byte of the column. In this way, since one column shift conversion is performed for every 4 bits, a figure having an inclination of Tan-1 (1/4) with respect to the original image is stored in the image buffer. Note that since no data is transferred to the lower left and upper right corners of the image buffer, which are indicated by diagonal lines in FIG. 2, the bits in these areas are set to 0 as the margins of the image.

Generally, slanted typefaces, such as italic characters, are T.
By applying a slope of an -1 (1/4) to the original image, the original slope of the font is removed.

Note that the slope given to this original image is Tan −1(
1/4), the inclination may be changed depending on the font selected by the font selection means 110, or the image data after the shift processing may be
The data may be shifted once more by displaying it on RT7 and having the operator instruct the inclination, so that the operator can confirm it.

In this way, before storing the input data in the image memory, the data can be shifted and the slanted fonts corrected for recognition. There is no need to keep it, and
This eliminates the possibility of adjacent data being included when cutting out characters.

In addition, so far we have described an example in which shifting processing is performed by dividing image data vertically, storing each column (vertical direction) in image memory, and storing one column's worth of data in a temporary buffer. It is also possible to perform the shifting process by dividing the data horizontally, storing it in the image memory row by row (horizontal direction), and storing one row's worth of data in a temporary buffer.

[Effects of the Invention] According to the present invention, by performing data shifting processing, it is possible to remove a slant from a slanted font in an original image, so that the following effects can be obtained.

(1) Characters can be cut out accurately.

(2) The recognition dictionary for characters with slanted fonts can be used as a recognition dictionary for characters with no slant. As a result, the memory capacity for the dictionary can be reduced.

(3) Since the slant is removed, it is possible to apply the normal recognition algorithm for characters without a slant as is.

[Brief explanation of the drawing]

Fig. 1 is a block diagram showing the first embodiment of the present invention Fig. 2 is a diagram explaining the shift processing in the first embodiment Fig. 3 is a diagram showing the number of boxes for normal characters and italic characters Figures 1 (A), 2 (A) and 3 (301) showing the configuration of a character recognition device.

Claims (2)

[Claims] (1) Instructing to shift the image information in a desired manner, storing the image information one column at a time, and when shifting the image information is instructed, shifting the stored image information. A character recognition method characterized by controlling to transfer. (2) an instruction means for instructing to shift the image information in a desired manner; a storage means for storing the image information one column at a time; and when the instruction means instructs to shift the image information; , a character recognition device comprising transfer control means for controlling the image information stored in the storage means to be shifted and transferred.