JPS60254284A - Character recognizer - Google Patents

Abstract

PURPOSE:To improve a character recognition factor by using the character presence signal of a binary coded signal to detect the slope of the character array on an original and correcting it. CONSTITUTION:The picture signal supplied from a scanner 13 is supplied to a preprocessing part 17 and converted into a binary coded signal to be stored to a picture memory 18. While the clock signal, character presence signal and main and secondary scan synchronizing signals are supplied to a slope detecting part 19 for calculation of the sloping angle of the character array on an original. Ad address generating part 20 produces the scan address of the memory 18 and the memory write signal according to the sloping angle of the character array given from the part 19 and an area designating parameter given from an operation part 21. The read signal read out of the memory 18 is written to a display memory 22 after correction of tis slope and then displayed at a display part 23. At the same time, a feature extracting part 25 extracts the feature quantity of a pattern equivalent to one character. This character pattern is recognized as character codes at a recognition part 26.

Description

DETAILED DESCRIPTION OF THE INVENTION (Field of Industrial Application) The present invention relates to a character recognition device that converts each character pattern read from a document into a character code.

(Structure of conventional example and its problems) 1- Conventionally, the optical image input device of a character recognition device that converts each character pattern read from a document 1 into characters and character codes has been By sequentially scanning the document 1 placed in the scanning area 2 while alternately repeating movement in the direction and movement in the sub-scanning direction, all characters 3 written on the document 1 are detected by the light receiving means. It is sequentially converted into an image signal. By the way, if the character arrangement of the document 1 is tilted with respect to the main scanning direction, the input character pattern will also be tilted, making extraction of the feature amount of the character pattern inaccurate, and lowering the recognition rate of the character pattern.

Therefore, the original 1 should be placed in the scanning area 2 so that the character arrangement of the original 1 is not tilted with respect to the main scanning direction, that is, the inclination angle α of the character arrangement with respect to the main scanning direction is O. It was very difficult to do so.

FIG. 2 shows the configuration of a conventional character recognition device equipped with means for correcting the inclination of the document 1 with respect to the main scanning direction, and 4 indicates movement of the reading position of the light receiving means in the main scanning direction. By repeating the movement in the sub-scanning direction 2- and sequentially scanning the original 1 placed in the scanning area 2 (see Figure 1), all characters 3 written on the original 1- are scanned. An optical image input bag W (hereinafter referred to as a scanner) 5 converts the image signal inputted from the scanner 4 into a binary code, and converts the image signal inputted from the scanner 4 into a binary code, and also converts the image signal of the original 1 in the main scanning direction. This preprocessing unit 5 detects the tilt angle β.
Original 1 when placed in scanning area 2 with backgrounds of different densities.
The inclination angle β of the document 1 is detected by detecting a boundary line that stands out on the edge of the document 1, or a specific symbol (not shown) for detecting inclination given in advance to the document 1. 6 is an image memory that stores the binary code output from the preprocessing unit 5; 7 is a display unit that displays a part of the content stored in the image memory 6; 8 is a slope output from the preprocessing unit 5; a character cutting unit that cuts out a character pattern for one character from the image memory 6 based on the detection signal and the area specifying parameter output from the operation unit 9;
Reference numeral 10 denotes a feature extraction unit that extracts the feature amount of the character pattern for one character output from the character extraction unit 8; This is a recognition unit that compares the character pattern with the content and recognizes the extracted character pattern as a character code.

In the conventional example configured in this way, the inclination angle β of the original 1 with respect to the main scanning direction is detected from the boundary line between the original 1 and the scanning area 2 or a specific symbol for inclination detection, and the character is detected with respect to the main scanning direction. Since the inclination angle α of the array is corrected to be O, and the feature amount of the character pattern is extracted, the recognition rate of the character pattern is decreased by -1.

However, when the character arrangement is tilted with respect to the edge of the document 1, a specific symbol for tilt detection, etc., as seen in the document 1 copied by a copying machine, etc., the document 1 may be Even if the inclination angle β becomes O, the inclination angle α of the character arrangement with respect to the main scanning direction does not become 0, so there is a drawback that the recognition rate of character patterns does not necessarily improve.

In addition, the method of adding a specific symbol for tilt detection to the original 1 in advance requires that the specific symbol for tilt detection be assigned accurately so that the tilt angle of the character arrangement with respect to the main scanning direction 4 times α becomes O. The work was troublesome because it did not work.

(Object of the Invention) An object of the present invention is to increase the recognition rate of character patterns by accurately detecting the inclination of a character arrangement written on a document from a binarized image signal.

(Structure of the Invention) The present invention divides a scanning area into a plurality of sub-areas in the sub-scanning direction from a binarized image input signal, and divides the scanning area into a plurality of sub-areas in the sub-scanning direction, and calculates the minimum coordinate value in the main-scanning direction of the character presence signal in each sub-area. Alternatively, a tilt detection unit is provided that detects a sub-region feature point having the maximum coordinate value and calculates the tilt angle of the character arrangement of the document using each detected sub-region feature point.

(Description of Embodiment) FIG. 3 shows the configuration of an embodiment of the present invention.
By sequentially scanning the document 15 placed in the 5-scanning area 14 while repeating the movement of the reading position of the light receiving means in the main scanning direction and the sub-scanning direction (see Fig. 4). ), an optical image input device (hereinafter referred to as a scanner) that sequentially converts all the characters 16 written on the document 15 into image signals using a light receiving means, and 17 converts the image signal input from the scanner 13 into a binary code. 18 is a preprocessing unit 17 which outputs a clock signal C1, a character presence signal D (synchronized with clock signal C), a main scanning synchronization signal H, and a sub-scanning synchronization signal V.
An image memory 19 stores the binary code outputted from the preprocessing unit 17, and an image memory 19 stores the characters of the document based on the clock signal C1, the character charge code D, the main scanning synchronization signal H, and the subscanning synchronization signal V output from the preprocessing unit 17. A tilt detection unit (the internal configuration and operation of which will be described later) calculates the tilt angle θ of the array (see FIG. 4), and 20 indicates the tilt angle θ of the character array of the document 15 output from the tilt detection unit 19. , the area specifying parameter output from the operation unit 21, that is, the area scanning start address X. ,
Yo, horizontal length A of the area, vertical length B of the area (fifth
6- An address generation unit that generates the scanning address MADD and memory write signal DM of the image memory 18 based on the area selection signal S (see Figure 4) and the area selection signal S (see Figure 4); When the scan address MADD of the image memo January 8 is output, the read signal RD is output from the image memory 18. Reference numeral 22 denotes a display memory that stores the content to be displayed on the display section 23, and a character pattern whose inclination has been corrected is written into this display memory 22 when the memory write signal DM is at the °'H'' level. The character pattern memory 24 stores a character pattern for one character, and a tilt-corrected character pattern is written into this character pattern memory 24 when the memory write signal DM is at the IIL'' level. 25 is an image memo January 8
A feature extraction section 26 extracts the feature amount of the character pattern for one character outputted from the reading signal RD and the character pattern memory 24, and the feature extraction section 26 extracts the feature amount output from the feature extraction section 25 and the contents of the dictionary 27 set in advance. This is a recognition unit that recognizes a given character pattern as a character code by comparing it with the character code.

Next, the specific configuration of the tilt detection section 19 will be explained with reference to FIG. In 191, the scanning area 147- of the scanner 13 is divided into four sub-areas in the sub-scanning direction, and the length of the first sub-area (■) in the sub-scanning direction is ρ1, and the length of the second sub-area (■) in the sub-scanning direction is ρ1. The length in the direction is Q2, the third sub area■
The length in the sub-scanning direction of the sub-scanning direction is Q3, and the length of the fourth sub-region (2) in the sub-scanning direction is preset as g4 (see FIG. 4). HI
+ level, count the main scanning synchronization signal H, (C,
−, the number of main scanning synchronization signals H≦(count value)≦Q,
number of main scanning synchronization signals H] (however, i = 1. to 4.
．． When No=O), the area selection signal S, (however, i
= 1 to 4) becomes the "'H" level, and the area selection signals S3. S2. The area selection unit 193 that outputs S3 or S4 determines that when the area selection signal S is at the IL HI+ level and the main scanning synchronization signal H is at the "T(" level), the character presence signal is at the 'H' level. The first point is detected for each main scan, and the minimum coordinate value point in the main scanning direction in the entire sub-region I is taken as the sub-region feature point, and the coordinate values (xl, Yl) of the sub-region feature point are detected. I minimum value detection unit,
194, when the area selection signal S2 is at the ``H'' level and the main scanning synchronizing signal H is at the ``H'' level, the first point at which the character presence signal becomes the II H I+ level is determined for each main scan. Minimum value detection unit, 195 , the area selection signal S3 is II H
At the II level, when the main scanning synchronization signal H is at the II H jl level, the first point where the character presence signal reaches the "H" level is detected for each main scanning, and the main scanning direction in the entire sub-area ■ is detected. The minimum value detection unit 196 detects the coordinate value (x a = y 3) of the sub-region feature point using the minimum coordinate value point as the sub-region feature point.
When the main scanning synchronizing signal H is at the °H" level, the first point where the character signal becomes the II)I II level is detected for each main scan, and the entire sub-area ■ The minimum value detection unit 197 detects the coordinate values (X4, Y4) of the sub-region feature point, with the minimum coordinate value point in the main scanning direction as the sub-region feature point. This is a calculation unit that linearly approximates (X+, Y+) [i=1 to 4] using the method of least squares, and calculates the inclination angle O of the character arrangement of the original 15 from the inclination angle of the straight line.

In this embodiment configured in this way, the inclination angle O is
Since it is generally small, there is no practical problem even if 4≦θ≦T. Now, when the tilt angle θ of the character arrangement of the original document 15 is output from the tilt detection unit 19, the area designation parameter X for displaying the image memo January 8 is output from the operation unit 21. t
The horizontal length A of the Y o l area, the vertical length B of the area, and the area selection signal S (S = "H" level) are output.Then, the address generation unit 20 executes the processing flow shown in FIG. Accordingly, starting from the coordinate point (x o , y o ), a corresponding address value in the image memory 18 is generated, and a read signal RD is written in the display memory 22. This is performed for the entire rectangular area shown in FIG. As a result, a part of the content of the image memory 18 is stored in the display memory 22 with the tilt corrected, so that the original 15 with a zero tilt is displayed on the display section 23.
is displayed. Next, from the operation unit 21, the image memory 18
Area parameter X for recognizing characters in . l Y
Given the horizontal length A of the O l area, the vertical length B of the area, and the area selection signal S (S="L" level),
Similarly, a character pattern for one character is read from the image memory 18, and the skew is corrected and stored in the character pattern memory 24. Therefore, the feature extractor 25 extracts the contents of the character pattern memory 24 or the read signal RD of the image memory 18.
It is possible to extract features of character patterns without considering the slope.

In this example, the minimum coordinate value point in the main operation direction was used as the sub-region feature point, but a method using the maximum coordinate value point or a straight line using the least squares method as a calculation between the coordinate values of the sub-region feature point In addition to approximation, there are also methods that use the average tilt angle between each two sub-region feature points, or remove sub-region feature points that deviate by more than a certain angle from the average tilt angle due to noise, and calculate the remaining sub-region feature points. It is also easy to consider a method of calculating the inclination angle using .

(Effects of the Invention) As explained above, according to the present invention, the inclination of the character arrangement of a document can be correctly detected using the character presence signal of the binarized image signal, and the character pattern with the inclination corrected can be detected. Since it is easily obtained, the recognition rate of the two-character recognition device can be improved. In addition, it is not necessary that all the character arrangement in each sub-area be in the correct format (if the characters start from the left edge of the document), but it is only necessary that at least one line be in the correct format, so there are no restrictions when detecting tilt. is extremely small and is particularly effective when reading printed originals.

[Brief explanation of drawings]

Fig. 1 is a diagram showing the scanning area of the scanner and the arrangement of the original, Fig. 2 is a block diagram of a conventional character recognition device, Fig. 3 is a block diagram of an embodiment of the present invention, and Fig. 4 is the arrangement of the original. FIG. 5 is an example of scanning an image memory when there is a tilt, and FIG. 6 is a diagram showing the configuration of a tilt detecting section used in an embodiment of the present invention. FIG. 7 is a processing flow diagram of the address generation section used in one embodiment of the present invention. 13... Optical image input device, 14... Scanning area,
15... Original, 19... Tilt detection unit, t, n, m
. ■...Sub area. Figure 2 Figure 3 CW Mother L Figure 5 Figure 6

Claims (1)

[Claims] In a character recognition device that converts a character pattern read from a document using an optical image input device into a character code,
dividing the scanning area into a plurality of sub-areas in the sub-scanning direction from the binarized image input signal, and having a minimum coordinate value or maximum coordinate value in the main-scanning direction among the character presence signals in each sub-area; 1. A character recognition device comprising: a tilt detection section that detects sub-region feature points in the document and calculates a tilt angle of a character arrangement in the document using the detected sub-region feature points.