JPH01124082A - Character recognizing device - Google Patents

Abstract

PURPOSE:To highly accurately segment and recognize a character by deciding a character or a character group on the basis of the width of a character block, and in case of a character group, segmenting each character by a polygon, extracting each prescribed feature variable, comparing the extracted result with a reference character pattern, and recognizing a character pattern with the shortest distance. CONSTITUTION:A character/character group deciding means 32 decides a single character when the width of a character block is small, and in the other case, decides a character group. A polygon tracing means 33 traces the contour of a character and calculates the relational parameters of an octagon surrounding the character to obtain the octagon. A character boundary correcting means 34 segments a character completely by an octagon, and when the segmented octagon satisfies a specific condition, a connected character processing means 35 recognizes the segmented character as a connected character. Since the length of respective sides of a polygon is automatically adjusted, a necessary character can be surrounded by a polygon having respectively different shapes, so that the accuracy of character segmentation and the recognition ratio of a character can be improved.

Description

[Detailed Description of the Invention] Industrial Application Field The present invention is a character recognition device that recognizes characters by first cutting out characters from a document with high precision using an electronic computer, and then performing matching calculations (verification calculations) on character patterns. It is related to.

Conventional technology In recent years, personal computers and power processors have become widespread in offices and households, resulting in so-called office automation.
ation) and home automation (Ho
It brought about the trend of automation. However, current personal computers and word processors still rely on data entry using keyboards. The system of inputting data using a gear board has the drawbacks that the input speed is slow and the operator is likely to feel tired. For this reason, there is a demand for the development of a high-speed and easy-to-use input method. The character recognition device according to the present invention was invented to be able to satisfy the above requirements.

FIG. 2 is a block diagram showing a conventional character recognition device. In the figure, 10, for example, scans a document with an image scanner and marks the white dots on the document with "o".
Binary data with the sunspot as “1”
20 is an image data input means for converting the data into a buffer memory section and storing it in a buffer memory section, and 20 inspects the binary data stored in the buffer memory section and detects the characters in each character sequence. This is a character string cutting means that inspects the position coordinates of a string, detects and records the position coordinates of each character. 30 is a character extraction means for detecting and recording binary data within the range of each character string, and 40 is a character extraction means for analyzing each of the extracted characters and, for example, calculating peripheral features thereof to extract features for identification. This is a feature extraction means for Reference numeral 55 denotes a standard character pattern in which the characteristics of each identification character, for example, its peripheral characteristics, are stored in a memory section for comparison with the input characters. Reference numeral 50 denotes a comparison between the character features from the feature extracting means 40 and all the character features from the standard character pattern 55 to find the distance between them, and the smallest of the distances is used as the recognition result. It is a means. 60 is an output means for displaying the recognition result on a display device such as a terminal display.

In a character recognition device, character segmentation is an extremely important preprocessing process, and if an error occurs during character segmentation, the subsequent identification will inevitably be inaccurate. The character extraction means 30 in the general character recognition device shown in FIG. There is something like the one described, but it uses a histogram of characters. As shown in Figure 3, first the amount of histogram projected on the horizontal axis of the character string is calculated, and the area where the amount of histogram exceeds the threshold value (threshold) indicates that there is a character, and the amount of histogram is calculated. The points where is below the threshold represent the separation points between characters. Also, for example, in Japanese Special Publication No. 60-53916,
A method of cutting out characters by calculating, comparing, and combining parameters of circumscribed rectangles in each element of the character is disclosed, and it is possible to cut out characters with separation parts such as the orthographic character "eight". However, it still has the disadvantage that it cannot accurately cut out italic type (1ta-lie) as shown in FIG. 4 or connected characters in which multiple characters are combined.

The problem to be solved by the invention is that the above-mentioned conventional character extraction means based on the histogram amount of characters is not reliable for the extraction of orthographic characters with large changes in the histogram amount between characters, that is, regular fonts. valid, but italic characters, e.g.
As shown in the figure, for italic characters or connected characters whose histograms overlap between characters, it becomes impossible to determine the separation point between the characters based on the amount of histogram. This is a fatal drawback of the method of extracting characters based on histogram quantities.

SUMMARY OF THE INVENTION In view of the above problems, it is an object of the present invention to provide a character recognition device that recognizes cut-out characters with high accuracy.

Means for Solving the Problems In order to solve the above problems, the present invention provides standard character patterns stored with predetermined feature quantities and two
An image data input means for converting into value data, a character string cutting means for cutting out a character string from the input image data, and a histogram amount projected perpendicularly to the direction of the character string to detect character boundaries. a character block boundary detection means; a determination means for determining whether the character block is a character or a character group consisting of a plurality of characters based on the width of the character block;
In the case of a group of characters, a polygonal character extraction means for cutting out each character using a polygon, a feature extraction means for extracting a predetermined feature amount from each cut-out character, the extracted feature amount and the above standard character, This character recognition device is characterized by comprising a comparison means, etc., which compares the character pattern with the character pattern and determines the character pattern with the smallest distance as the recognition result.

According to the present invention, the character recognition device configured as described above can recognize characters regardless of whether it is an italic character, an orthographic character, or a concatenated character.
Characters can be extracted with high precision, and the recognition rate of the entire character recognition device can be improved.

Example FIG. 1 is a block diagram showing one embodiment of the present invention.

In FIG. 1, parts that perform the same operations as those in FIG. 2 are given the same reference numerals. In FIG. 1, lO is
An image data input means for scanning a document and converting it into video binary data, 2° is a character string cutting means for cutting out a character string by adjusting the histogram amount and threshold for the vertical axis, and 31 is a character string cutting means for cutting out a character string from left to right. This is a character block boundary detection means that calculates the amount of histogram projected on the horizontal axis along the character string, and detects whether a threshold for determining a character block has been exceeded.

The histogram amount can be calculated using the following formula.

It is assumed that the character string exists between the coordinates 'Yl( and yL) (Figure 5).

Here, H(X) indicates the histogram of point X1 projected on the horizontal axis, and p (x+, y) is the pixel (X1).

y) is the binarized value of the image. If the black point is ``1°'' and the white point is ``O'', then the condition for detecting a character block boundary is when the histogram amount is '°0''° or when the right end of the line has already been reached. .

Reference numeral 32 denotes a character or character group determination means for determining whether what is included in a character block is a single character or a character group consisting of a plurality of characters. The criteria for this judgment are as follows.

That is, if the width of the character block is smaller than (Xav, +N, +ze), it is determined that it is a single character; otherwise, it is determined that it is a group of characters. Here, X i
v a is the average width of the extracted characters [its initial value (i
N*+xe indicates the width of the noise (according to empirical values).

33 is a polygon tracking means. In this embodiment, a polygon with the largest side is used. The tracking operation scans the character block from left to right, finds the black dot on the leftmost side of the first character, and then traces the outer edge of the character using this black dot as the starting point. As shown, the relational parameters of the octagon surrounding the character are calculated to obtain the octagon. The width and height of the octagon are detected, and if both are smaller than a predetermined value, the characters within this hexagon are discarded as noise. Next, calculations similar to those above are repeated until the entire block of characters has been processed. FIG. 7 shows the calculation of parameters related to the above octagon.

Each side of this octagon is defined as follows.

Side 1: X ” 2x-y=C4 Above X ITla X s X 1.l+ 11,'
l m t x, 'I+n+n, C1, C2, C3,
By calculating the values of and C4, we can draw an octagon surrounding the character.The width of this hexagon is
X-xlnll, and the height is y, nAX-ylnll.

It is.

34 is a character boundary correction means, and this means corrects characters such as "i", "j", "'%", '"-" and some Kanji characters having two or more separated parts. It can be completely cut into hexagons. The correction method is the 8th
As shown in the figure, if the frame of the "original hexagon" in FIG. If a black dot exists, the black dot is used as the starting point, and the outer periphery of the black dot is traced again using the hexagon tracing means 33 described above, and it is stored in the original hexagon to form the character. Boundaries can be corrected.

35 is a connected character processing means for cutting out two or more consecutive characters.

If the cut out hexagon satisfies the following five conditions, it is recognized as a connected character.

(1), (x+nax-xmtn) > (xavex
7/4) (2), Vm+n > (j tap +
5) (3), (C3C1)> (XavaX 2
)(4), (2X Xmjn C2)> (Y
mtn+ Vmax)/2(5)0 The histogram amount near the center line of the character block is (VmaXV+n+
n) Less than X3/4.

Here, X n v e is the average width of the extracted characters, t 1lar is the starting position of the character string, and X □
88, X m t n s ylns x s V m
+ n, C1, C2, C3, and C4 are all the same as the definitions described for the hexagonal tracking means 33 above.

The method for cutting out connected characters is to first calculate the histogram amount of the connected characters on the horizontal axis, select the one with the smallest histogram amount, and then divide the connected characters in the diagonal direction along the diagonal line (x+3y=c). In this method, the histogram amounts are determined, the smallest one is selected, and then both obtained histogram amounts are compared and the smaller point is used as the break point.

Since one character is enclosed by the variable polygon formed by each of the above means, this polygon becomes a circumscribed polygon of the character. 9th
The hexagon shown in the figure has the length and position of each side adjusted according to the shape of the font. For characters with different fonts and sizes, the character recognition device of the present invention automatically adjusts the side lengths of the polygons to enclose the necessary characters with polygons of different shapes. I can put it out. An example of successful extraction of an italic character and a combined character is shown in FIG. What I would like to emphasize here is that the character extraction function of this invention is not only effective for characters in various fonts (orthographic or italic), but also for connected characters or characters that are partially separated. is also valid.

A hardware configuration diagram of the character recognition device of this embodiment is shown in FIG. In FIG. 11, a recognition board (Recognitio board) at the center of the character recognition device according to the present invention is shown.
n Board) is equipped with a digital signal processor (Digital Signal Processor).
or) is included. The main operations of the recognition process are executed by the processor. After obtaining binary image data of the document to be recognized using an image scanner, this data is inspected and character strings or characters are cut out by the above-described processing. Next, as parameters representing the characteristics of the extracted characters, the boundary density and background density shown in the Proceedings of the National Conference of the Japan Institute of Electronics and Communication Engineers (1985, P1577) are known. This feature is calculated, and each character is divided into small sections to obtain a feature amount of a predetermined number of bytes. This is compared point by point with feature quantities of the same number of bytes in the corresponding section of a standard character pattern created in advance in the same way, and the character pattern with the smallest distance is taken as the recognition result. In the case of this embodiment, the standard character pattern includes multiple types of fonts, and clustering (cluster-i)
ng) technique can be used to support multi-font recognition while saving the storage space required to store the entire standard character pattern.

This invention is not limited to the above-mentioned embodiments (and can be implemented with appropriate modifications as long as the gist is not changed. For example, the polygon surrounding the characters is not limited to an octagon, Similar segmentation results can be obtained using decagons or hexagons.

Further, the cutting target of the present invention is not limited to characters, but can also be applied to figures, and the accuracy of the cutting is not affected in any way.

Further, the data input method is not limited to using an image scanner, and image data can also be stored in the memory section by using an input method using a TV camera or the like.

Results of the Invention According to the present invention, the above-mentioned character recognition device is able to cut out characters using polygonal arithmetic, and to recognize 100 characters such as uppercase and lowercase English letters, numbers, and special symbols. By calculating the characteristics of the density and background density, the characteristics can be extracted, and the accuracy rate of character extraction is 99.8% or more, and the character recognition rate is 98.9% or more. As a result, the practicality is much greater.

figure

[Brief explanation of the drawing]

FIG. 1 is a block diagram showing the configuration of an embodiment of the character recognition device of the present invention, FIG. 2 is a block diagram showing the configuration of a conventional character recognition device, and FIGS. 3 and 4 are histograms in the conventional example. Figure 5 is an explanatory diagram showing an example of a character string, Figure 6 is an explanatory diagram showing the definition of octagonal parameters in an embodiment of the present invention, and Figure 7 is an octagonal diagram. Figure 8 is an explanatory diagram showing the correction area of character boundaries, Figure 9 is an explanatory diagram showing the expansion and contraction direction of each side of an octagon, and Figure 10 is an octagon with connected characters. FIG. 11, an explanatory diagram showing an example of successful extraction of italic characters, is a hardware layout diagram in an embodiment of the present invention. (lO)...image data input means, (20)...
... Character string extraction means, (31) ... Character block boundary search means, (32) ... Character or character group judgment means, (33) ... Octagon tracing means, (34) ...・Character boundary correction means, (35)...Concatenated character processing means, (
40)... Feature extraction means, (50)... Comparison means,
(55)...Standard character pattern, (60)...Output means. Name of agent: Patent attorney Toshio Nakao

Claims (3)

[Claims] (1) A standard character pattern stored with predetermined feature quantities, an image data input means for scanning an image and converting it into binary data, and a character string cutting means for cutting out a character string from the input image data; character block boundary detection means for detecting character boundaries using a histogram amount projected perpendicularly to the character string direction; a determining means for determining, a polygonal character cutting means for cutting out each character using a polygon if it is a group of characters, a feature extracting means for extracting a predetermined feature amount for each cut out character, and an extracted feature amount. A character recognition device comprising a comparison means for comparing the character pattern and the standard character pattern described above and determining a character pattern having the smallest distance therebetween as a recognition result. (2) The character recognition device according to claim 1, wherein the polygonal character cutting means includes means for determining a frame of each character pattern. (3) The character according to claim 2, wherein the polygonal character cutting means includes character separation part correction means for including the separated part of each character in the polygon in each cut out character. recognition device. (4) The polygonal character cutting means determines that it is a connected character in which multiple characters are connected according to the polygonal condition, and extracts the most of the histogram amount in the vertical or diagonal direction with respect to the character string direction. 3. The character recognition device according to claim 2, further comprising connected character processing means for separating connected characters by using smaller portions as break points.