JPS6262389B2 - - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION The present invention relates to a method and apparatus for automatically reading characters in documents such as arbitrary forms and books.

Conventional devices of this kind have a readable document with a reference mark printed in advance, or a specified frame with characters written correctly in it, etc., depending on its format and format. However, there was a drawback that it was impossible to read characters from arbitrary documents such as forms and books.

In order to eliminate the above-mentioned conventional drawbacks, the present invention provides a form,
A document such as a book is made into a blurred image, the character string is patterned, the blurred image is scanned to recognize the position, length, etc. of the character string, and individual characters are read according to the recognition results. I did it like this,
The purpose is to provide a method and apparatus for automatically reading characters from any document accurately and in a short time. Hereinafter, the invention will be explained in detail with reference to the drawings.

FIG. 1 is a schematic configuration diagram showing an embodiment of the automatic character reading device of the present invention, in which 1 is a document such as an arbitrary form or book, 2 is an imaging section, 3 is a blur image creation section, and 4
9 is a character string recognition unit; 9 is a character reading means for reading each character in the character string according to the recognition result of the character string; Consists of 8. Then,
When the blurred image creation unit 3 issues a command to the imaging unit 2 to shift its focus from the document 1, the imaging unit 2 captures a blurred image of the document 1 with the focus shifted,
The blurred image is binarized and patterned by the blurred image creation section 3, the character string recognition section 4 recognizes the position, length, etc. of the character string, and the reading head drive section 5 performs the recognition based on the recognition result. The reading head 6 is scanned along the character strings of the document 1 to extract character information for each row, and the character information is cut out one character at a time by a character cutting section 7, recognized by a character recognition section 8, and coded. It's becoming like that.

FIG. 2 shows the configuration of the imaging section 2, and in the figure,
21 is a TV camera whose shooting range covers the entire document 1;
22 is a focus adjustment mechanism for the TV camera 21; 23 is a belt; 24 is a pulley; 25 is a step motor;
Reference numeral 6 indicates information on the amount by which the focus position of the TV camera 21 is moved from the document 1, and 27 indicates a motor drive circuit that determines and operates the rotation angle of the step motor 25 based on the information 26. Here information 26 is document 1
This information is given manually or from a control circuit (not shown) based on the document size, character size, and feedback information on the recognition results of the character string recognition unit 4. When a command to shift the focal point position is given from the blurred image creation section 3, the motor drive circuit 27 rotates the step motor 25 by a predetermined angle based on the information 26, and rotates the pulley 24. The rotation of the pulley 24 is transmitted to the focus adjustment mechanism 22 via the belt 23, and the focal position of the TV camera 21 is shifted by a predetermined amount. The blurred image photographed in this state is sent to the blurred image creation section 3.

FIG. 3 is a block diagram of the binarization circuit of the blurred image creation section 3. In a blurred image of characters, the density of characters with high density such as kanji and characters with low density such as kana and numbers are significantly different, and if the entire image is binarized with the same threshold, the characters This is inappropriate because a blur pattern that accurately indicates the rows cannot be obtained. Therefore, as shown in FIG. 4a, the image memory 31 storing the blurred image is divided into a number of small areas 31a, a density histogram as shown in FIG. 4b is taken for the small areas 31a, and its inflection point A is calculated. An appropriate blur pattern can be obtained by binarizing as a threshold value and performing this for each small area. In FIG. 3, reference numeral 31 denotes an image memory with intermediate density, which stores the blurred image sent out from the imaging section 2. 32 is a small area 31a of the image memory 31
33 is a histogram creation circuit that creates a density histogram of the small area 31a based on the information in the buffer memory 32; 34;
35 is an inflection point detection circuit that detects an inflection point from the histogram; 36 is a threshold; and 37 is a comparator. To explain the operation, first, the small area 31a is taken out from the image memory 31 and stored in the buffer memory 32.
Based on this, a density histogram is created by the histogram creation circuit 33 and stored in the memory 34.
The threshold value 36 is determined by the inflection point detection circuit 35 from the density histogram in the memory 34, and the threshold value 36 is determined by the buffer memory 32.
The comparator 37 compares the information with the information of the small area 31a in the image memory 31, converts it into a binary value, and stores it in the original location in the image memory 31. Thereafter, similar processing is performed on all other small areas of the image memory 31 to convert the blurred image into a binary blurred pattern and send it to the character string recognition section 4.

Figures 5a and 5b show an example of the original character pattern and its blur pattern, respectively, and the respective density distributions along the scanning line S. In the binarized blur pattern, high frequency components included in the original pattern are shown. You can see that it has been removed.

By the way, a blurred image obtained by shifting the focus in the developing section 2 has a different field of view from an image captured in a focused state. For this reason, it is necessary to correct the field of view when recognizing character strings. FIG. 6 shows the visual field correction circuit 38 of the blurred image creation section 3, which receives information 26 on the amount by which the focal position of the TV camera 21 is moved from the document 1, and shows the corresponding converted amount 39 of visual field correction. is calculated and sent to the character string recognition section 4.

FIG. 7 is a block diagram of the character string recognition unit 4.
4 is an image memory that stores the blurred image sent from the blurred image creation unit 3; 42 is a character area extraction circuit that extracts only the character area excluding "appendices, drawings", etc. from the blurred image; 43 is an image memory 4
1, a tilt correction circuit that detects the tilt of the image in 1 and corrects it by rotating the TV camera 21, etc.; 44, a m,σ calculation circuit that calculates the average value m of the width of the blur pattern of the character string and its standard deviation σ; 45; 46 is a character string line extraction circuit that extracts a character string line by averaging the midpoints of the blur pattern of a character string, and 46 is a character string end detection circuit that detects the upper, lower, left, and right ends of the blur pattern of a character string. Character area extraction circuit 42 and tilt correction circuit 4
3 is unnecessary when the object to be read is a document consisting only of characters, and when the document is not tilted due to a document guide or the like.

FIG. 8 shows an example of a binarized blur image stored in the image memory 41, in which 101 is a character string blur pattern, 102 is the width of the blur pattern 101, 103 is the midpoint of the width 102, 104
105 is the break in the blur pattern 101, 105 is the line feed part of the blur pattern 101, 106 is the constriction of the blur pattern 101, and 107 is the blur pattern 1
01 is unevenness on the edge, 108 is noise patterned by dirt on the paper surface, 109, 110, 111, 1
12 is a character string line obtained by averaging the coordinates of the center point 103 of the blur pattern 101, 113 and 114 are arrows representing the X coordinate parallel to the character string and the Y coordinate perpendicular to the character string, and 115 is a scanning line in the Y coordinate direction. be.

Next, the operation of character string recognition will be described. first,
The m,σ calculation circuit 44 scans the entire screen in the direction of the scanning line 115, and calculates the width 102 of the blur pattern 101.
The average value m and its standard deviation σ are calculated and sent to the character string line extraction circuit 45. Next, the character string line extraction circuit 45 performs scanning with the y direction as the main scan for each blur pattern, determines the coordinates of the midpoint of the blur pattern according to the flowchart shown in FIG. Determine the coordinates of the column lines. now,
The coordinate l(k-1) of the character string line 109 has already been obtained, and the coordinate l(k) of the character string line 110 is newly calculated.
When trying to find the , the scan start command is used to scan in the y direction with the y coordinate of the character string line 109 as the starting point, and the middle point C(n) of the blur pattern that first intersects (n is the scan in the x direction). ) and the width d are calculated. In judgment 1, a judgment is made to remove the unevenness 107 and noise 108 at the edge based on the width d and m, the average value m and standard deviation σ obtained by the σ calculation circuit 44, and if Yes, the C(n) and d are After erasing, scanning in the y direction is continued, and if NO, proceed to judgment 2. In judgment 2, it is judged whether or not the cut 104 has been passed using the coordinates C(n-1) of the midpoint obtained during the previous scan, and if Yes, the coordinates C(n-1) of the midpoint are used.
(n) and d are deleted to start the next scan, and if NO, proceed to decision 3. In judgment 3, it is judged whether or not the line feed part 105 or the constriction 106 has been passed, and if NO, the character string line 11 required by C(n) is
It is determined that it belongs to the blur pattern of 0, this is stored, and the next scan is started, and in the case of Yes, the process proceeds to determination 4. Judgment 4 uses the y-coordinate l(k-1) of the previous character string line 109 to determine whether it is the line break 105 or the constriction 106 that has passed, and if Yes, it is determined that it is the constriction 106. Then that C
(n), erase d and continue scanning in the y direction,
If NO, it is determined that the line feed section 105 has been passed and C
It is determined that (n) is the first midpoint belonging to the desired character string line 110, the coordinates of the previous midpoint are deleted, and the next scan is performed. When scanning in the y direction is completed to the right end of the screen, the midpoint coordinate C(n) is determined to belong to the blur pattern of the character string line 110.
The y-coordinate l(k) of the character string line 110 is calculated by averaging. Similarly, the coordinates below the character string line 111 can be determined.

As shown in FIG. 10, the character string end detection circuit 46 performs vertical scanning 116, 117 from the left end to the right end of the character string line 110 using the coordinates of the character string line, for example 110, as a reference, to detect the blur pattern. The maximum value in the vertical direction is detected, the coordinates of the upper end 118 and the lower end 119 are detected, and horizontal scanning 120, 121 is performed with a predetermined width centered on the character string line 110 to detect the maximum value in the horizontal direction of the blur pattern. The coordinates of the left end 122 and right end 123 are detected.

The recognition results of the position, length, and width of the character string obtained as described above are sent to the reading head drive unit 5, and the reading head 6 is driven as described above. will be done. In addition,
Regarding the reading head drive section 5, reading head 6, character cutting section 7, and character recognition section 8, those used in conventional character reading devices can be used as they are, so no special mention will be made of them. Further, the reading head drive section 5,
TV camera 21 of the imaging section 2 instead of the reading head 6
may be scannable and zoomable and controlled to read individual characters of a string.

As explained above, according to the method of the present invention, a document such as a form or a book is made into a blurred image, the character string is patterned, and the blurred image is scanned to recognize the position, length, etc. of the character string. Since each character is read in accordance with the recognition result, characters can be read accurately no matter what format or form the document is to be recognized. Further, according to the device of the present invention, an imaging unit captures a blurred image of a document such as a form or a book by shifting its focal position from the document surface;
a blurred image creation unit that exchanges the blurred image with a binarized and patterned image and corrects changes in the field of view of the imaging unit due to movement of the focal position;
Scan the patterned image and determine the position of the character string,
It consists of a character string recognition unit that recognizes the length, etc., and a character reading means that reads each character in the character string according to the recognition results such as the position, length, etc. of the character string.
By simply binarizing the image obtained by moving the focal position of the camera, a patterned blurred image can be created, and the high-frequency components of the characters can be removed from the character string. Since it is a binary pattern, recognition by scanning is accurate and the scanning circuit is also simplified, and since a conventional character reading device can be used as is as a character reading means for reading each character, the cost of the device can be reduced. In addition, conventionally, the contents of documents with arbitrary formats and formats were manually input into various processing devices using keyboards, etc., but with this device, it is possible to read them as is, so the input processing time is reduced. It has the advantage that it can be used as an automatic input device when creating a database from a large amount of existing documents and as an input/output device for a terminal in data communication.

[Brief explanation of the drawing]

The drawings show embodiments of the present invention, and FIG. 1 is a schematic diagram of the automatic character reading device of the present invention, FIG. 2 is a diagram of the configuration of the imaging section, and FIG. 3 is a diagram of the configuration of the binarization circuit. Figure 4a is an explanatory diagram of division of a blurred image;
Figure b is the density histogram of a small area of the blurred image.
Fig. 5a is an explanatory diagram of the original pattern of characters and its density distribution, Fig. 5b is an explanatory diagram of the blurred pattern of characters and its density distribution, Fig. 6 is an explanatory diagram of the visual field correction circuit, and Fig. 7 is an explanatory diagram of the characters. FIG. 8 is an explanatory diagram showing an example of a blurred image; FIG. 9 is a flowchart of a character string line extraction circuit; and FIG. 10 is an explanatory diagram of detecting the top, bottom, left, and right edges of a blur pattern. . DESCRIPTION OF SYMBOLS 1...Document, 2...Imaging unit, 3...Blurred image creation unit, 4...Character string recognition unit, 9...Reading head drive unit 5, reading head 6, character cutting unit 7, character recognition unit 8 A means of reading characters.

Claims (1)

[Scope of Claims] 1 A document such as a form or book is made into a blurred image, the character string is patterned, the blurred image is scanned to recognize the position, length, etc. of the character string, and the recognition result is obtained. An automatic character reading method characterized in that individual characters are read. 2. An imaging unit that captures a blurred image of a document such as a form or book by shifting its focal position from the document surface, and an imaging unit that converts the blurred image into a binarized and patterned image and that a blur image creation unit that corrects changes in the visual field of the imaging unit; a character string recognition unit that scans the patterned image and recognizes the position, length, etc. of the character string; and a character string recognition unit that scans the patterned image and recognizes the position, length, etc. of the character string. 1. An automatic character reading device comprising character reading means for reading individual characters in a character string according to a recognition result of the character string.