JPH06150055A - Character recognizing device - Google Patents

Abstract

PURPOSE:To enable highly accurate recognition even for a document mixing Japanese and English sentences by judging whether the sentence is the Japanese sentence or English sentence while calculating a non-average height rate as the rate of a rectangle provided with a height higher than the average height of integrated circumscribed rectangles. CONSTITUTION:A recognizing area detection part 3 detects the rectangles circumscribed on the linked black picture elements of image data stored in an image data storage part 2 and detects an area for each attribute such as a character, graphic, table and image (photograph) basing on the size of circumscribed rectangles, the density of black picture elements and the complicated degree of picture elements. On the other hand, an attribute sort part 4 sorts the respective areas corresponding to the attributes so as to perform recognition processing corresponding to the respective attributes. Then, concerning the are where the character is sorted by attribute by the attribute sorting part 4, a Japanese/English sentence judging means 5 judges whether that area is a Japanese sentence area or an English sentence area basing on the features of characters themselves inside the area. In this case, the Japanese sentence/English sentence is judged while calculating the non-average height rate as the rate of the rectangle having the height higher than the average height of integrated circumscribed rectangles.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention captures a document image by using an optical means such as a scanner, detects an area for each attribute such as a character, a figure and a table from the captured image data, and recognizes according to each attribute. The present invention relates to a character recognition device that performs processing.

[0002]

2. Description of the Related Art In recent years, digitization of documents and the like has been widely performed for the purpose of creating a database of printed documents and reusing documents, and a character recognition device is used as an input device to a computer or the like for that purpose. .

A conventional character recognition device will be described below. In order to improve the recognition efficiency, the conventional character recognition device distinguishes a character area detected from image data into a Japanese sentence area and an English sentence area, and performs character recognition processing separately.

In order to determine whether the character area is a Japanese text area or an English text area, there are two types of spaces in an English sentence, that is, a space between words and a space between characters in a word, and a rectangle circumscribing a character. When the distribution of the distances between the two is detected, the peaks of the distribution can be detected at two points. However, in Japanese, only the spaces between the characters exist. Therefore, if the distribution of the distances between the rectangles circumscribing the characters is taken, It was based on the feature that no two peaks were detected.

The operation of the character recognizing device constructed as above will be described below. First, the rectangle circumscribing the black pixels connected from the image data captured by the scanner is detected as the circumscribed rectangle, and the character circumscribed rectangle and the circumscribed rectangle other than the character are determined from the size of the circumscribed rectangle and the black pixel density in the circumscribed rectangle. Classify into.

Next, the character area is detected by integrating the classified character circumscribing rectangles. Next, the circumscribed rectangle of the specified character area is detected.

Next, the circumscribed rectangles that are adjacent to the left and right of each of the detected circumscribed rectangles are extracted and the distribution of the distances between the circumscribed rectangles is obtained.

Next, when two peaks of the distribution of the distance between the circumscribed rectangles can be detected, the character area is determined to be an English text area, and the other areas are determined to be a Japanese text area.

Next, the Japanese area recognizes Japanese characters such as kanji, hiragana and katakana, and the English area recognizes alphabetic characters such as alphabets.

[0010]

However, in the above-mentioned conventional configuration, in a monospace document created by a typewriter or the like, the distance between circumscribing rectangles changes depending on the type of character, and a document written in italics or the like is used. Since the fonts are tilted and the circumscribed rectangles overlap, the space between characters in the word could not be detected stably.

Therefore, even in the English text area, the distribution of the distances of the circumscribed rectangles does not have two peaks, and the English text area is erroneously determined as the Japanese text area, and the recognition rate is lowered, resulting in reduced reliability and versatility. It had a problem of chipping.

The present invention solves the above-mentioned problems of the prior art. Character recognition excellent in reliability and versatility that can obtain an accurate and stable recognition result regardless of the type of document and the type of font. The purpose is to provide a device.

[0013]

In order to achieve this object, a character recognition device of the present invention detects a character area which detects a coordinate of a character area recognized from a binarized recognition target document from a connected black pixel feature. Means, a circumscribing rectangle detecting means for detecting a circumscribing rectangle of connected black pixels in the character area detected by the character area detecting means, and detecting a total number of circumscribing rectangles, and a circumscribing rectangle overlapping another circumscribing rectangle. A circumscribing rectangle unifying unit that detects a circumscribing rectangle that circumscribes those circumscribing rectangles, an overlap ratio calculating unit that calculates a ratio of a circumscribing rectangle that overlaps another circumscribing rectangle, and an upper and lower overlap that calculates a ratio of a circumscribing rectangle that vertically overlaps. A ratio calculating means, a circumscribing rectangle average height calculating means for calculating the height of the circumscribing rectangle having the highest appearance frequency as the circumscribing rectangle average height, and a circumscribing rectangle flat calculated by the circumscribing rectangle average height calculating means. A non-average height ratio calculating means for calculating a ratio of a circumscribed rectangle having a height equal to or higher than a height, an overlapping ratio calculated by the overlapping ratio calculating means, and an upper and lower overlapping ratio calculated by the upper and lower overlapping ratio calculating means A Japanese-English sentence determining unit that determines whether the character region is a Japanese sentence region or an English sentence region based on the non-average height percentage calculated by the average height percentage calculating unit, and a Japanese sentence region determined by the Japanese sentence English sentence judging unit. And a Japanese character recognizing unit for recognizing Japanese characters, and an English character recognizing unit for recognizing English characters in the English sentence region judged by the Japanese sentence English sentence judging unit.

[0014]

[Operation] With this configuration, in order to determine whether the text area is Japanese or English based on the characteristics of Japanese and English characters, it is possible to accurately determine whether the text area is a Japanese area or an English area regardless of the document creation method or typeface. It is possible to make a determination and improve the recognition accuracy.

[0015]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS An embodiment of the present invention will be described below with reference to the drawings.

FIG. 1 is a functional block diagram of a character recognition device according to an embodiment of the present invention, and FIG. 2 is a device block diagram of the character recognition device.

Reference numeral 1 denotes an image data capturing section for converting a recognized image document into image data binarized by a scanner and capturing the image data. Reference numeral 2 denotes an image data storage section for storing the image data captured by the image data capturing section 1. Detects a rectangle circumscribing the concatenated black pixels of the image data stored in the image data storage unit 2, and detects characters, figures, tables, images (based on the size of the circumscribing rectangle, the density of black pixels, and the pixel complexity). Recognition area detection unit 4 for detecting an area for each attribute of (photograph), 4 is an attribute selection unit for performing selection processing according to each attribute so as to perform recognition processing according to each attribute, and 5 is an attribute selection unit 4 Is a Japanese-English sentence determination unit that determines whether the region is a Japanese sentence region or an English sentence region based on the characteristics of the characters themselves in the region, and 6 is an attribute selection item. An image (photo) processing unit that processes a region whose attribute is selected from the image (photo) in the unit 4, and 7 is a Japanese character such as Kanji, Hiragana, or Katakana when the determination result by the Japanese-English sentence determination unit 5 is a Japanese region. For recognizing characters for Japanese characters, 8 for recognizing characters for English characters such as alphabets when the result of judgment by the Japanese-English sentence judging unit 5 is an English region, 9 for attribute selection unit 4 This is a table / graphics processing unit that processes areas whose attributes are tables and graphics.

In FIG. 2, 10 is a central processing unit (hereinafter abbreviated as CPU) that controls the entire character recognition device, 11 is a read-only memory (hereinafter abbreviated as ROM) that stores a character recognition program, and 12 is an image. Random access memory (hereinafter abbreviated as RAM) in which data and data of a recognition program are stored, 13 is a scanner that reads a document to be recognized and converts it into image data, and 14 is a CPU 10 from outside to start and end recognition. Keyboard for giving commands, 15 is RAM by CPU10
12 is a CRT that displays the recognition result and the like stored in 12.

The Japanese / English sentence determination means 5 determines whether the region is a Japanese region or an English region based on the feature of the character itself. The feature of the English character used at that time is (1) capital letter "A" , There are variations in the height of the characters, such as the lower case “a”. (2) Since there are few separated characters, when a circumscribed rectangle of connected black pixels is detected, one circumscribed rectangle corresponds to one character. (3) The circumscribed rectangle rarely overlaps with other circumscribed rectangles. In the case of italicized characters, the circumscribing rectangle overlaps with other circumscribing rectangles, but even in this case, the circumscribing rectangle does not overlap in the vertical direction even though it may overlap in the horizontal direction.

On the other hand, the characteristics of Japanese characters are: (1) There is no variation in character height as seen in English characters. (2) Japanese characters are often formed by a plurality of circumscribing rectangles so that the Kanji characters are composed of one-sided and one-sided. (3) The circumscribing rectangles that form Japanese characters overlap vertically, horizontally, and vertically. And so on.

The operation of the Japanese / English sentence determining means of the character recognizing device configured as described above will be described below with reference to FIGS.
This will be described with reference to 2.

FIG. 3 is a flow chart of the initialization process of the Japanese / English sentence determination means of the character recognition device in this embodiment, FIG. 4 is a flow chart of the circumscribed rectangle integration process of the Japanese / English sentence determination means, and FIG. 5 is a Japanese / English determination. 6 is a flowchart showing a concrete example of image data, FIG. 7 and FIG. 10 are diagrams showing a concrete example of a circumscribed rectangle detection processing result, and FIG. FIG. 11 is a diagram showing a specific example of the circumscribed rectangle integration processing result,
FIG. 12 is a diagram showing a specific example of the circumscribing rectangle detection processing.

First, initial setting processing is performed. In FIG. 3, first, the circumscribed rectangles of the connected black pixels in the designated character area are detected, and the coordinate information of all the circumscribed rectangles is stored in the RAM 12 (S1). Here, in the coordinate system of the coordinates of the circumscribed rectangle, the upper left corner is the origin, the horizontal coordinate is x, and the vertical coordinate is y, as shown in FIG. The circumscribed rectangle is represented by the upper left coordinates (x1, y1) and the lower right coordinates (x2, y2). In the example of circumscribing rectangle detection for the English image data in FIG. 6, eleven circumscribing rectangles K1 to K11 are detected.
Further, an example of circumscribing rectangle detection of Japanese image data in FIG. 9 is shown in FIG.
At 0, 11 circumscribing rectangles K12 to K22 are detected. The work area laywork of the RAM 12 is cleared to 0 to detect the average height of the circumscribing rectangle (S2).
Next, the storage order of the circumscribed rectangles is rearranged in the ascending y1 order in order to reduce the number of times the circumscribed rectangles are overlapped (S3).
Here, the circumscribed rectangles in FIG. 7 are sorted by y1, so that K1 → K2 → K5 → K11 → K6 → K9 → K3 → K.
They are stored in the order of 4 → K7 → K8 → K10. Also, FIG.
The circumscribed rectangle is K12 → K15 → K18 → K13 → K1
4 → K19 → K21 → K22 → K16 → K17 → K20
In that order. Next, recTotal that stores the total number of detected circumscribing rectangles, mCt that stores the number of circumscribing rectangles that overlap other circumscribing rectangles, and v that stores the number of circumscribing rectangles that vertically overlap.
mCt is set to 0 (S4).

Next, circumscribing rectangle integration processing is performed. In FIG. 4, first, in order to detect the total number of circumscribed rectangles and inspect the overlap of the circumscribed rectangles, the first circumscribed rectangle sorted in S3 is set in bsBox storing the address of the circumscribed rectangle to be inspected (S5). . Here, the circumscribed rectangle of K1 is set in FIG. 7, and the circumscribed rectangle of K12 is set in FIG. next,
The circumscribing rectangle total number counter recTotal is incremented by 1 (S
6). The circumscribing rectangle address next to bsBox is set in objBox that stores the address of the target circumscribing rectangle for checking whether or not there is a circumscribing rectangle overlapping with bsBox (S
7). Here, if bsBox is K1 in FIG. 7, obj is
Box is K2, and if bsBox is K12 in FIG. 10, K15 is set in objBox.

Next, there is an objBox, and obj
It is checked whether the y1 value of the Box is smaller than the y2 + 1 value of the bsBox (S8). bsB if no
The circumscribing rectangle overlapping ox is jumped to S16 as not existing after objBox. If yes, bs
As there is a possibility that a circumscribed rectangle that overlaps the Box exists, it is checked whether bsBox and objBox overlap (S9). Here, x1 of bsBox is objBox
X2 of bsBox is smaller than objB
If it is larger than x1 of ox, it is determined that bsBox and objBox overlap. In Figure 9, bsBox is K1
The case where the objBox is K13 in 2 corresponds to this example.
If no, jump to S15. If yes, a circumscribed rectangle number counter m that overlaps another circumscribed rectangle
Advance Ct by 1 (S10).

Next, it is checked whether or not bsBox and objBox are vertically overlapped (S11). Where bsBox
X2 of x and the smaller value of x2 of objBox and bsBo
If the difference between x1 of x and the larger value of x1 of objBox is greater than the difference between the smaller value of y2 of bsBox and y2 of objBox and the larger value of y1 of bsBox and y1 of objBox, bsBox And objBo
It is determined that x overlaps vertically. In FIG. 9, bsB
This is the case when ox is K18 and objBox is K17. If no, jump to S13. ye
If it is s, the circumscribed rectangle number counter vm that overlaps vertically
One Ct is recommended (S12).

Next, a rectangle circumscribing two circumscribing rectangles bsBox and objBox is detected, and the coordinates thereof are set to bsB.
The circumscribed rectangles are integrated by resetting to ox (S1
3). The data of objBox is deleted (S14). Here, as an example, bsBox in FIG. 9 is objB with K18.
When ox is K17, the circumscribed rectangle of K18 and K17 is shown in Fig. 1.
At T14 of 1, the − coordinate of T14 is set in bsBox, and K17 of objBox is deleted. objBox
Next data of is reset to objBox and jumped to S8
p (S15). S8 to S1 until there is no circumscribed rectangle to be inspected that overlaps with the current bsBox.
The processes up to 5 are repeated.

Next, if it is determined that the circumscribing rectangle that overlaps with bsBox no longer exists, the current target b
It is determined that the integration of sBox is completed, and the height number of bsBox in the layWork area is incremented by 1 (S16). Here, the height of bsBox is bsBox
Y2-bsBox y1 + 1. As an example, the height of the integrated circumscribing rectangle T12 in FIG. 11 is calculated to be 12, and the twelfth count of layWork is advanced by one. bsBo currently targeted for bsBox
The circumscribed rectangle data next to x is set (S17). next,
It is checked whether data exists in bsBox (S18). y
If it is es, jump to S6, and from S6 to S17
The process of is repeated. If it is no, the circumscribing rectangle integration process is terminated.

By the above processing, rectotal,
The values of mCt, vmCt, and layWork are set.

For the English image data in FIG. 6, re
ctotal = 11, mCt = 0, vmCt = 0, la
5th in yWork = 4, 6th in layWork =
9th of 1, layWork = 3, and for the Japanese image data of FIG. 9, rectotal = 11, m
Ct = 5, vmCt = 1, 11th of layWork =
3, 12 = 3 in layWork.

Next, a Japanese / English sentence determination process is performed. In FIG. 5, a circumscribed rectangle number counter mCt that overlaps another circumscribed rectangle
And the ratio of the circumscribing rectangle number counter vmCt that vertically overlaps with the total circumscribing rectangle number recTotal is calculated (S1).
9). Here, the overlap ratio and the top-bottom overlap ratio are respectively the overlap ratio = mCt / rectotal × 100.
%, Overlapping ratio = vmCt / rectotal * 1
It is calculated as 00%. As an example, with respect to the English image data in FIG. 6, the overlapping ratio = 0, the upper and lower overlapping ratio =
0, for the Japanese image data in FIG. 9, the overlapping ratio = 45.4, and the vertical overlapping ratio = 9.0.

Next, the average height of the integrated circumscribing rectangles is detected (S20). The position having the highest numerical value in the layWork is the circumscribing rectangle average height. When there are a plurality of positions with the maximum numerical values in the layWork, the one with the largest position is the circumscribing rectangle average height. As an example, for the English image data of FIG. 6, the fifth laywork is the highest at 4, so the average height of the circumscribed rectangle is 5, and for the Japanese image data of FIG. The third and twelfth are 3 and the same number, but the one having the larger position is adopted and the circumscribed rectangle average height is set to 12.

Next, the number of circumscribing rectangles having a height of 1.25 times or more the average height of the circumscribing rectangles among the integrated circumscribing rectangles is calculated, and the ratio to the total circumscribing rectangles recTotal is calculated as a non-average height ratio. Is calculated as (S21). As an example, there are five integrated circumscribing rectangles in FIG. 8 having a height of 1.25 times the circumscribing rectangle average height 5, and the non-average height ratio is 45.5%. Further, in the circumscribing rectangle in which the figure 11 is integrated, 0 is 1.25 times or more the circumscribing rectangle average height 12, and the non-average height ratio is naturally 0%.
Is.

Next, it is checked whether or not the overlapping ratio is equal to or larger than the first threshold value (= 20) set in advance (S22).
If no, jump to S25. If yes, the upper and lower overlapping ratio is greater than or equal to the second threshold (= 10)
Is checked (S23). If yes, the area is determined to be Japanese and jump to S27. If no, it is checked whether the upper and lower overlapping ratio is equal to or more than a third threshold value (= 1/4) of the overlapping ratio (S24). ye
If it is s, the area is judged to be Japanese and jump to S27. If no, the non-average height ratio is
The fourth threshold times (= 2/3 times) of the overlapping ratio is changed to the fifth threshold (=
It is checked whether it is larger than the value obtained by adding 10) (S25). n
If it is o, the area is judged to be Japanese and jump to S27. If yes, the area is determined to be English, the attribute of the area is set to English, and all processing is terminated (S26). Next, the area is determined to be Japanese, the attribute of the area is set to Japanese, and all the processing ends (S27).

As a result of the above processing, the image data of FIG. 6 is determined to be an English region because the overlapping ratio is 0%, the vertical overlapping ratio is 0%, and the non-average height ratio is 45.5%. The image data is judged to be a Japanese sentence area because the overlapping ratio is 45.4%, the vertical overlapping ratio is 9.0%, and the non-average height ratio is 0%. You can judge correctly whether there is.

[0036]

As described above, according to the present invention, when the circumscribing rectangles are integrated, the overlapping ratio, which is the ratio of the circumscribing rectangle that overlaps the other circumscribing rectangle, and the vertical overlapping ratio, which is the ratio of the circumscribing rectangles that vertically overlap, are calculated. Then, calculate the non-average height ratio, which is the ratio of rectangles that have a height equal to or higher than the average height of the integrated circumscribing rectangle, and pay attention to the characteristics of the Japanese and English characters themselves obtained from this information. Since Japanese / English judgment is performed by using the Japanese / English judgment, it is possible to judge the proposed Japanese / English sentence regardless of the type of document such as monospaced document or italic typeface, and it is highly accurate even for documents with mixed Japanese / English sentence. It is possible to realize a character recognition device which is highly reliable and versatile for recognition.

[Brief description of drawings]

FIG. 1 is a functional block diagram of a character recognition device according to an embodiment of the present invention.

FIG. 2 is a device block diagram of a character recognition device according to an embodiment of the present invention.

FIG. 3 is a flowchart of an initial setting process of Japanese-English sentence determination means of the character recognition device in the embodiment of the present invention.

FIG. 4 is a flowchart of a circumscribing rectangle integration process of the Japanese / English sentence determination means of the character recognition device in the embodiment of the present invention.

FIG. 5 is a flowchart of a Japanese / English sentence determination process of a Japanese / English sentence determination unit of the character recognition device in one embodiment of the present invention.

FIG. 6 is a diagram showing a specific example of image data according to an embodiment of the present invention.

FIG. 7 is a diagram showing a specific example of a circumscribed rectangle detection processing result according to an embodiment of the present invention.

FIG. 8 is a diagram showing a specific example of a circumscribed rectangle integration processing result according to an embodiment of the present invention.

FIG. 9 is a diagram showing a specific example of image data according to an embodiment of the present invention.

FIG. 10 is a diagram showing a specific example of a circumscribed rectangle detection processing result according to an embodiment of the present invention.

FIG. 11 is a diagram showing a specific example of a circumscribing rectangle integration processing result according to an embodiment of the present invention.

FIG. 12 is a diagram showing a specific example of circumscribing rectangle detection processing according to an embodiment of the present invention.

[Explanation of symbols]

 1 Image Data Importing Section 2 Image Data Storage Section 3 Recognition Area Detection Section 4 Attribute Selection Section 5 Japanese / English Judgment Section 6 Image (Photo) Processing Section 7 Japanese Character Recognition Section 8 English Character Recognition Section 9 Table / Graphic Processing Section 10 CPU 11 ROM 12 RAM 13 Scanner 14 Keyboard 15 CRT

Claims (1)

[Claims] 1. A character area detecting unit for detecting coordinates of a character area recognized from a binarized recognition target document from a connected black pixel feature, and a connected black area for a character area detected by the character area detecting unit. A circumscribing rectangle detecting means for detecting a circumscribing rectangle of pixels to detect the total circumscribing rectangle, a circumscribing rectangle unifying means for detecting a circumscribing rectangle overlapping another circumscribing rectangle and detecting a circumscribing rectangle circumscribing the circumscribing rectangle, and the like. Of the circumscribing rectangle that overlaps with the circumscribing rectangle of the circumscribing rectangle, the overlapping ratio calculating means of calculating the ratio of the circumscribing rectangle that vertically overlaps, and the height of the circumscribing rectangle with the highest appearance frequency is the circumscribing rectangle average. A circumscribing rectangle average height calculating means for calculating the height, and a non-average height ratio calculation for calculating a ratio of circumscribing rectangles having a height equal to or higher than the circumscribing rectangle average height calculated by the circumscribing rectangle average height calculating means. And a character area based on the overlapping ratio calculated by the overlapping ratio calculating unit, the vertical overlapping ratio calculated by the vertical overlapping ratio calculating unit, and the non-average height ratio calculated by the non-average height ratio calculating unit. Is a Japanese-language area or an English-language area, a Japanese-English sentence determination means, a Japanese character recognition means for recognizing Japanese characters in the Japanese-language area determined by the Japanese-language sentence determination means, and a determination made by the Japanese-language sentence determination means. A character recognition device comprising an English character recognition means for recognizing an English character in a specified English text area.