JPH0822507A - Document recognition device - Google Patents

Abstract

PURPOSE:To provide the device whose recognition precision is hardly affected by noise by discriminating a ruled line area as it is and discriminating the scanner read noise as the noise to delete it from area information. CONSTITUTION:A CPU is provided with a picture input part 7, a picture reduction part 8, a connected black picture element rectangle must generation part 9, a ruled line attribute giving part 10, a character attribute giving part 11, a photograph attribute giving part 12, a table attribute giving part 13, a ruled line attribute giving part 14, a read noise detection part 15, 8 graphic attribute giving part 16, and 8 recognition processing part 17. A connected black picture element rectangle list detected from reduction picture data is generated, and attributes are set to connected black picture element rectangles by the ruled line attribute giving part 10, the character attribute giving part 11, the photograph attribute giving part 12, the table attribute giving part 13, the ruled line attribute giving part 14, and the graphic attribute giving part 16. The read noise detection part 15 discriminates whether a rectangle to which any attributes are not given is the scanner read noise or not, and rectangle information of the rectangle discriminated as the scanner read noise is deleted from the connected black picture element rectangle list.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention captures a document image of a printed document in which characters, tables, figures, and ruled lines are mixed by using an optical means such as a scanner, and character blocks and figures based on the captured image data. The present invention relates to a document recognition device that divides an area into blocks and the like and performs recognition according to the attribute of each block.

[0002]

2. Description of the Related Art A conventional document recognition process will be described below.

<Step 1> The binary data read by the scanner is reduced so that the resolution becomes about 100 DPI.

<Step 2> With respect to the reduced image data, the coordinates of the connected black pixel rectangle and the number of connected black pixels are stored in the memory.

<Step 3> From the connected black pixel rectangles extracted in step 2, rectangles forming a character are extracted.
If the shorter one of the width or height of the connected black pixel rectangle is smaller than a predetermined threshold value, the attribute of the character is given to the connected black pixel rectangle.

<Step 4> It is judged whether or not the connected black pixel rectangle which has not become a character attribute in step 3 is a rectangle having a photograph attribute. With respect to the connected black pixel rectangle, if a minute rectangle exists in the rectangle with a predetermined threshold value or more, or if the ratio of black pixels in the rectangle is more than the predetermined threshold value, the connected black pixel rectangle Is assigned a photo attribute.

<Step 5> It is checked whether the remaining connected black pixel rectangles that have not become characters / photographs in steps 3 and 4 are table rectangles. Check whether there is a horizontal line or a vertical line in the rectangle, and have a table attribute from the number of horizontal or vertical lines existing in the rectangle, the number of intersections of the horizontal line and the vertical line, and the black pixel information in the rectangle. Judge as a rectangle.

<Step 6> A graphic attribute is given to the connected black pixel rectangles to which no attribute has been given.

<Step 7> By the above processing, attributes of characters, photographs, tables and figures are given to each rectangle.
For rectangles having attributes of photographs, tables, and figures other than characters, one rectangle is considered to form one area, and photographs, tables,
A region having a graphic attribute is extracted.

<Step 8> Regarding a rectangle having a character attribute included in a photograph, a table or a figure, since the rectangle is likely to be a part of the figure, it is included in the photograph, the table or the figure. The rectangle with the character attribute is deleted from the black pixel rectangle list.

<Step 9> Character strings are integrated with respect to the rectangle having the character attribute remaining in step 8, and the character area is extracted.

<Step 10> The character area is subjected to character recognition, the image area is subjected to image compression, the table area is subjected to table recognition, and the graphic area is subjected to vectorization to obtain a recognition result corresponding to each attribute.

[0013]

However, in the prior art, since the area formed by a plurality of ruled lines and the scanner reading noise become a graphic area, the internal characters also become a part of the graphic. There was a problem of vectorization.

Therefore, an object of the present invention is to provide a document recognition device in which the recognition accuracy is less likely to be affected by noise.

[0015]

A document recognition apparatus according to the present invention comprises means for reducing image data of a binarized character recognition target document to obtain reduced image data, and black pixels from the reduced image data. The connected black pixel rectangle circumscribing the connected black pixel is detected, the coordinates of the connected black pixel rectangle corresponding to the reduced image data and the number of connected black pixels are stored as a connected black pixel rectangle information list. Information storage means,
A ruled line is one in which the aspect ratio of the connected black pixel rectangle is greater than or equal to a predetermined threshold, and a ruled line attribute is added as one of the connected black pixel rectangle information. A unit that determines a character rectangle that is less than or equal to a predetermined threshold value to give a character attribute to the connected black pixel rectangle information, a unit that gives a photograph attribute from the characteristics of the black pixels in the connected black pixel rectangle, and a connected black pixel Check whether there are parts that make up horizontal lines and vertical lines in the rectangle, and check the number of horizontal lines or vertical lines existing in the connected black pixel rectangle and the number of intersections of the horizontal line and vertical line, and the black pixels in the connected black pixel rectangle. Judging from the information as a table,
Target based on means for giving a table attribute to the connected black pixel rectangle information, horizontal line / vertical line information, the total number of black pixels in the connected black pixel rectangle, and the number of connected black pixels detected by the connected black pixel rectangle information storage means It is checked whether the rectangle is a rectangle composed of a plurality of ruled lines, and if so, means for giving the attribute of the ruled line to the rectangle information, and further, the rectangle is determined from the connected black pixel rectangle position information and the black pixel information inside the rectangle. It is provided with a means for checking whether there is image reading noise, and for deleting the information of the connected black pixel rectangle from the rectangle list when there is reading noise, and a means for giving a graphic attribute to the remaining rectangles to which no attribute is given. .

[0016]

With the above structure, the present invention can determine a ruled line area as a ruled line area, and an area where a plurality of ruled lines that are determined as a figure in the prior art are connected, and image reading noise. It can be determined and deleted from the area information.

[0017]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS An embodiment of the present invention will be described with reference to the drawings. FIG. 1 is a block diagram of a document recognition apparatus according to an embodiment of the present invention. In FIG. 1, reference numeral 1 denotes a central processing unit (hereinafter abbreviated as CPU) for recognizing a document, which is shown in FIG.
An image input unit 7, an image reduction unit 8, a connected black pixel rectangle list creation unit 9, a ruled line attribute assignment unit 10, a character attribute assignment unit 11, and a photograph attribute of (a functional block diagram of a document recognition apparatus according to an embodiment of the present invention). It includes an assigning unit 12, a table attribute assigning unit 13, a ruled line attribute assigning unit 14, a reading noise detecting unit 15, a graphic attribute assigning unit 16, and a recognition processing unit 17. 2 is a read-only memory (hereinafter, RO
M), and the random access memory 3 (hereinafter,
Image data read by the scanner 4 is stored in a RAM (abbreviated as RAM). Reference numeral 5 is a keyboard for giving a command to the CPU 1 from the outside, and 6 is a display device for displaying the recognition result recognized by the CPU 1.

The area division will be described below with reference to FIGS. 2 and 3 (image data exemplification diagram in one embodiment of the present invention) and FIG. 7 (flow chart of the document recognition apparatus in one embodiment of the present invention). . The image input unit 7 stores the image data captured by the scanner 4 in the RAM 3 (s1). The upper left corner of the image data is the origin, the horizontal coordinate is x, and the vertical coordinate is y. The image reduction unit 8 reduces the captured binary image data so as to have a resolution of 100 DPI in order to speed up attribute determination, and R
It is stored in AM3 (s2).

In the connected black pixel rectangle list creation unit 9, RA
From the reduced binary image data of M3, connected black pixel rectangles of black pixels connected in the vicinity of 8 are detected to create a connected black pixel list (s3). The data of the connected black pixel rectangle registered in the connected black pixel list is the upper left coordinates (x1, y1), the lower right coordinates (x2, y2), and the number of black pixels. In this embodiment, as a result of processing the image data shown in FIG.
It is assumed that the connected black pixel rectangle shown in FIG. 4 (an exemplary view of the connected black pixel extraction result in the embodiment of the present invention) is obtained.

In the ruled line attribute assigning section 10, the aspect ratio of the connected black pixel rectangle is set to a predetermined threshold value Th_RTO_L.
It is checked whether it is INE or more (s4). If the aspect ratio of the rectangle is greater than or equal to the threshold value Th_RTO_LINE, the rectangle is determined to have the ruled line attribute, and the process proceeds to s12.

With respect to the connected black pixel rectangle that has not become a ruled line in s4, the character attribute assigning unit 11 sets the length of the short side of the rectangle to a predetermined threshold value Th_MAX_CHAR or less, and the black pixel occupies the area. The ratio is the threshold Th_R
It is checked whether there is more than TO_CHAR (s5). Threshold Th_MAX in which the length of the short side of the rectangle is predetermined
If it is less than or equal to _CHAR and the ratio of black pixels in the area is greater than or equal to the threshold Th_RTO_CHAR, the rectangle has a character attribute, and the process proceeds to s13. Rectangle 22 of FIG.
23 becomes a rectangle having a character attribute and advances to s13, and advances to s5 as a rectangle other than the other rectangular characters.

Here, the black pixel density can be calculated by the following calculation. Black pixel density (d1) = black pixel number of connected black pixel rectangle / (rectangle width × rectangle height) × 100 (%) In the photo attribute assigning unit 12, for the rectangles remaining as rectangles other than ruled lines and characters Then, it is checked whether or not it is a rectangle having a photo attribute (s6). A threshold value Th_RTO_IM in which a ratio of a minute rectangle to a rectangle is predetermined
When it is G or more and the number of minute rectangles is equal to or more than a predetermined threshold value Th_CNT_IMG, it is determined that the rectangle has a photograph attribute, and the process proceeds to s14. The rectangle 26 in FIG. 4 is a rectangle having a photo attribute.

The table attribute assigning section 13 checks whether the rectangle has the table attribute. First, a line component detection process is performed to determine whether there is a line component in the rectangle. The method of line component extraction is to check whether the length of the black pixel is equal to or greater than the threshold value Th_LEN_LINE in the horizontal direction / vertical direction,
The table is judged based on the detected line component (s7). The threshold value Th_LEN in which the length of the detected horizontal line is the width of the rectangle
_TABLE times the threshold Th_CNT_TABLE
The number of (= 3) or more, and the length of the vertical line is equal to or more than the threshold value Th_LEN_TABLE of the height of the rectangle, the threshold value Th
There are more than _CNT_TABLEs, and a line crossing any of the above lines is a threshold Th_CRS_TAB.
If it is LE or more, the rectangle becomes a table. In the image data example of FIG. 3, the rectangle 24 of FIG. 4 is determined as the table. The rectangles determined to be the table proceed to s15, and the rectangles not determined to the table proceed to the second ruled line determination processing s8. The rectangle 20 in FIG.
Does not satisfy the conditions in the table because the line components are horizontal line 1 and vertical line 1.

The ruled line attribute assigning unit 14 performs ruled line determination as described below for rectangles that have not become a table (s8). s
It is determined from the relationship between the connected black pixel rectangle black pixel density detected in 3 and the total black pixel density in the target rectangle whether it is a ruled line area. The total black pixel density can be calculated by counting all the black pixels in the rectangle, dividing the number by the area of the rectangle, and multiplying by 100.

Total black pixel density (d2) = total number of black pixels in rectangle / (rectangle width × rectangle height) × 100 (%) The detected total black pixel density d2 is twice the black pixel density d1. As described above, there is a horizontal line detected by the table attribute assigning unit 13 that is Th_LEN_TABLE times the width of the rectangle, or a vertical line Th_LEN_TABL that is the height of the rectangle.
If there is E times or more, the rectangle becomes a ruled area rectangle. If the ruled line rectangle is determined, the process proceeds to s16, and if the ruled line is not determined, the process proceeds to the scanner reading noise detection process of s9. Since the rectangle 20 in FIG. 4 satisfies the above condition, a ruled line attribute is added, and the rectangle 21 in FIG. 4 does not become a ruled line.

The reading noise detecting section 15 determines whether or not the rectangle for which an attribute has not been set up to now is scanner reading noise (s9). The rectangle 21 in FIG. 4 is determined to be scanner reading noise, and the process proceeds to s17. In s17, the rectangle information determined as the scanner reading noise is deleted from the linked black pixel rectangle list.

The graphic attribute assigning section 16 assigns a graphic attribute to a rectangle whose attribute has not been determined based on the above determination criteria (s10). When it is given, it is recognized for each attribute (s11), and the process is ended.

Through the above processing, any attribute of ruled lines, characters, figures, photographs, and tables is added to all connected black pixel rectangles. A rectangle to which a ruled line, a figure, a table, or a photograph attribute is added becomes one area as it is. A rectangle having a character attribute included in the figure area is determined to be a part of the figure and deleted from the linked black pixel rectangle list. For example, for the image data example of FIG. 5 (image data exemplary diagram in one embodiment of the present invention),
Character attributes are given to rectangles other than the rectangle k12 in FIG. 6 (exemplary diagram of the connected black pixel extraction result in the embodiment of the present invention), but actually, the rectangles k6 to k9 are part of the figure. The same processing is performed on the existing photograph area as the figure area. A character string is extracted from the remaining character rectangle, and a character area is extracted based on the character string information.

In the recognition processing section 17, for each area obtained as described above, in the case of a character area, character cutting processing is performed and then character recognition processing is performed. In the case of the graphic area, the graphic is vectorized, and in the case of the table area, the structure of the table is recognized, and the character recognition processing is performed on each cell. Image compression is applied to the photo area to reduce the amount of information.

In this embodiment, the threshold value Th_MAX_C is set.
The values of HAR to Th_CNT_TABLE are as follows.

Th_RTO_LINE = 25 Th_MAX_CHAR = 100 Th_RTO_CHAR = 15 Th_RTO_IMG = 0.5 Th_CNT_IMG = 80 Th_LEN_LINE = 25 Th_LEN_TABLE = 4/5 Th_CNT_BBLE = S3

[0032]

According to the present invention, means for reducing image data of a binarized character recognition target document to obtain reduced image data, and detecting a portion where black pixels are connected from the reduced image data are detected. A connected black pixel rectangle that circumscribes the connected black pixel, and stores the connected black pixel rectangle coordinates corresponding to the reduced image data and the connected black pixel number as a connected black pixel rectangle information list; Means for determining a ruled line when the aspect ratio of the rectangle is equal to or greater than a predetermined threshold value, and assigning a ruled line attribute as one of the connected black pixel rectangle information;
Photographs based on the means for determining the character rectangle when the length of the side of the connected black pixel rectangle is less than or equal to a predetermined threshold and giving the character attribute to the connected black pixel rectangle information, and the characteristics of the black pixels in the connected black pixel rectangle. A means for giving an attribute and a part of the connected black pixel rectangle forming a horizontal line and a vertical line are checked, and the number of horizontal lines or vertical lines existing in the connected black pixel rectangle and the number of intersections of the horizontal line and the vertical line, and , Means for determining a table from the information of the black pixels in the connected black pixel rectangle and giving the table attribute to the connected black pixel rectangle information, horizontal line / vertical line information, and the total number of black pixels in the connected black pixel rectangle Based on the number of connected black pixels detected by the black pixel rectangle information storage means, it is checked whether the target rectangle is a rectangle composed of a plurality of ruled lines, and if so, means for giving a ruled line attribute to the rectangle information, The connected black pixel rectangle position information and the inside of the rectangle Check if the rectangle is image reading noise from the pixel information, and if there is reading noise, delete the information of the connected black pixel rectangle from the rectangle list, and set the graphic attribute for the remaining rectangles to which no attribute is given. Since it is provided with a means for providing, a region where a plurality of ruled lines that are conventionally determined to be a figure and image reading noise can be determined to be a ruled line region and a scanner reading noise is determined to be noise. Since it is deleted from the information, a character area surrounded by ruled lines, which was impossible in the past, can be accurately extracted as a character area without being integrated into the ruled line area, and highly accurate document recognition can be performed. .

Since the scanner reading noise is detected not when the data is read but when the attribute of the character or the like is determined, the copy noise can be deleted even for the document having the copy noise in the document to be recognized. Is.

[Brief description of drawings]

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

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

FIG. 3 is an exemplary diagram of image data according to an embodiment of the present invention.

FIG. 4 is a view showing an example of a result of extraction of connected black pixels according to an embodiment of the present invention.

FIG. 5 is a view showing an example of image data according to an embodiment of the present invention.

FIG. 6 is a view showing an example of a result of extraction of connected black pixels according to an embodiment of the present invention.

FIG. 7 is a flowchart of a document recognition device according to an embodiment of the present invention.

[Explanation of symbols]

 1 CPU 2 ROM 3 RAM 4 scanner 5 keyboard 6 display device 7 image input unit 8 image reduction unit 9 linked black pixel rectangle list creation unit 10 ruled line attribute addition unit 11 character attribute addition unit 12 photo attribute addition unit 13 table attribute addition unit 14 Ruled line attribute assigning unit 15 Reading noise detecting unit 16 Graphic attribute assigning unit 17 Recognition processing unit

Claims (1)

[Claims] 1. A means for reducing image data of a binarized character recognition target document to obtain reduced image data, and a portion where black pixels are connected from the reduced image data is detected to detect a connected black. A connection information storage unit that obtains a connection black pixel rectangle circumscribing a pixel and stores the coordinates of the connection black pixel rectangle corresponding to the reduced image data and the number of connection black pixels as a connection black pixel rectangle information list, and a vertical and horizontal direction of the connection black pixel rectangle. Means for determining a ruled line having a ratio equal to or greater than a predetermined threshold value and giving a ruled line attribute as one of the connected black pixel rectangle information, and a side length of the connected black pixel rectangle is equal to or less than a predetermined threshold value. To determine a character rectangle as a character rectangle and to add a character attribute to the connected black pixel rectangle information, a means to add a photo attribute from the characteristics of the black pixels in the connected black pixel rectangle, and a horizontal line and a vertical line in the connected black pixel rectangle. Parts that make up a line Whether there is a horizontal line or a vertical line in the connected black pixel rectangle, the number of intersections of the horizontal line and the vertical line, and the black pixel information in the connected black pixel rectangle. A means for adding a table attribute to the information, the horizontal line / vertical line information, the total number of black pixels in the connected black pixel rectangle, and the number of connected black pixels detected by the connected black pixel rectangle information storage means A method for checking whether the rectangle is composed of a plurality of ruled lines, and if so, adding a ruled line attribute to the rectangle information, and further reading the image from the rectangle based on the connected black pixel rectangle position information and the black pixel information inside the rectangle It is provided with means for checking whether it is noise, deleting the information of the connected black pixel rectangle from the rectangle list when it is read noise, and means for giving a graphic attribute to the remaining rectangles to which no attribute is given. Characteristic Document recognition device that.