JPH0433080A - Method for recognizing character in table - Google Patents

Abstract

PURPOSE:To enable character recognition by exactly extracting not only the line of laterally describing characters but also the line of longitudinally describing characters in a table by discriminating the lengths of each frame in main scanning and sub scanning directions and switching the method of extracting the character line in each frame corresponding to the result of the discrimination. CONSTITUTION:The frame surrounded by line segments in the main scanning and sub scanning directions is extracted from the table area of a document picture. Then, the lengths of each frame in the main scanning and sub scanning directions are discriminated and corresponding to the discriminated result, the method for extracting the character line in each frame is switched. Namely, it is automatically discriminated whether the character printed in each frame of the table is described laterally or longitudinally, and the character line extraction method suitable for the discriminated direction is applied. Thus, even in the table mixing the laterally and longitudinally describing frames, the character line in each frame is exactly extracted, and the accuracy for the character recognition is improved.

Description

DETAILED DESCRIPTION OF THE INVENTION [Field of Industrial Application] The present invention relates to a method for recognizing characters within a document in a character recognition device.

[Conventional technology]

In character recognition devices, five document images are often divided into a character area, an image area such as a photograph or figure, a table area, etc., and different processing is performed on each area.

Regarding the table area, it recognizes the position of the borders that make up the table,
Connected black pixels are traced in an image within a frame surrounded by ruled lines, a circumscribed rectangle of connected black pixels is found, and the circumscribed rectangles are integrated to extract character lines and character recognition is performed.

(Problem to be solved by the invention) However, conventionally, character line extraction processing is performed on the assumption that the characters in each frame of a table are printed either horizontally or vertically. In the case of a table with a mixture of horizontal and vertical text lines as shown in Figure 3, if horizontal text (vertical writing) is assumed, the vertical text lines (
(Horizontal character lines) may not be extracted accurately, and as a result, character recognition may not be performed accurately.

An object of the present invention is to provide a table character recognition method that can accurately extract and recognize both horizontally written character lines and vertically written character lines in a table.

[Means to solve the problem]

The present invention provides a table character recognition method that extracts frames surrounded by line segments in the main scanning direction and sub-scanning direction from a table area of a document image, extracts character lines within each frame, and recognizes the characters. Determine whether the character line in each frame is written horizontally or vertically based on the length in the main scanning direction and the length in the sub-scanning direction of each frame,
The method is characterized in that the method for extracting character lines within each frame is switched depending on the result of this determination.

[For production]

According to the present invention, it is automatically determined whether the characters printed in each frame in the table are written horizontally or vertically, and a character line extraction method suitable for the determined direction is applied. Therefore, even in tables with a mixture of horizontal and vertical frames,
The character lines within each frame are accurately extracted, thus increasing the accuracy of character extraction and character recognition within each frame.

〔Example〕

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

FIG. 2 is a flowchart of the process.

A document is read by a binary image input unit 101 such as a scanner, and the binary image is stored in the binary image memory 102 (processing step 201).

For this document image, the table area recognition unit 103 automatically recognizes the table area using run length distribution or the like.
Alternatively, an area specified by the operator using a mouse or the like is recognized as a table area, and an image of the table area is stored in the table area image memory 104 (processing step 202).

For this image of the table area, the line segment extraction unit 1 in the main scanning direction
05, black pixels connected in the main scanning direction are traced to extract a line segment in the main scanning direction, and the coordinates of the starting point and end point are stored in the main scanning direction line segment coordinate memory 106 (processing step 203). Similarly, the sub-scanning direction line segment extraction unit 107 traces the black pixels connected in the sub-scanning direction in the table area image, extracts a line segment in the sub-scanning direction, and calculates the coordinates of the starting point and end point along the sub-scanning direction line. The coordinates are stored in the minute coordinate memory 108 (processing step 204).

Next, in the frame recognition unit 109, each memory 106° 108
The frame of the table is recognized by the combination of the line segment in the main scanning direction and the line segment in the sub-scanning direction, and the coordinates of the frame, for example, the coordinates of the diagonal vertices, are stored in the frame coordinate memory 110 (
Process step 205).

In addition, in the frame area extraction unit 111, the frame coordinate memory 115
By referring to the frame coordinates in , the image of the frame area is extracted from the table area image memory 104 and stored in the frame area image memory 112 (processing step 206).

Next, in the row direction determination unit 113, the frame coordinate memory 110
A histogram of lengths in the main scanning direction and the sub-scanning direction is created for all frames with reference to (processing step 207
, 208), and all frames with a length in the sub-scanning direction of & large degrees are determined to be frames with horizontal writing in the line direction. The remaining frames are determined to be frames with vertical writing in the line direction (processing steps 211 and 212), and the remaining frames are determined to be vertical frames with the largest number of frames determined up to that point (processing steps 211 and 212). 213), and processing steps 207,
When calculating the histogram in step 208, a certain amount of width is given to the length in each scanning direction.

Similarly, when determining the length in processing step 211, if the difference in length to be compared is within a certain width range, it is determined that they match. The obtained row direction information is sent to the circumscribed rectangle extraction unit 11
4 to the row image extraction unit 116.

For example, in the case of the table shown in FIG. 3, the lengths of the horizontally written frames in the sub-scanning direction are all the same (or almost the same), so the frequency is maximum. Therefore, all horizontal writing frames in this table are determined to be horizontal writing in processing step 210. Also, the length of the frame in which the vertically written character string "data" in this table is printed in the main scanning direction is almost the same as the length of the maximum frequency in the sub-scanning direction (the difference is within a certain width range) So.

In processing step 212, it is determined that the frame is for vertical writing.

Next, the circumscribed rectangle extraction unit 114 refers to the frame area image memory 112, traces the connected black pixels for the image within each frame, extracts the circumscribed rectangle of connected black pixels, and extracts the circumscribed rectangle of the connected black pixels, and The coordinates are stored in the circumscribed rectangle coordinate memory 115 (processing step 214).

Next, the row image extraction unit 116 refers to the circumscribed rectangle coordinate memory 115 and extracts a circumscribed rectangle connected to black pixels for each frame area image in the frame area image memory 112 by the line direction determination unit 11.
By integrating in the row direction determined by 13, character row images (character string images) within the frame are extracted and stored in row image memory 1.
17 (processing steps 215, 216) - In this way, the line direction for each frame is determined, that is, whether it is horizontal writing or vertical writing, and character line image extraction is performed using a method suitable for the determined line direction. Therefore, in a table area where horizontal writing frames and vertical writing frames coexist, it becomes possible to accurately extract character line images in any row direction.

Next, in the character recognition unit 118, a character image is cut out from each type of character line image in the line image memory 117, since the character line image extraction in the previous stage is accurate.

This character image cutting out can also be performed accurately.

Then, the features of the cut out character image are extracted, recognized by matching with the recognition dictionary, and the recognition result is output to the outside (processing step 217).

〔Effect of the invention〕

As explained above, according to the present invention, it is automatically determined whether a character line printed in each frame in a table is written horizontally or vertically, and the character line is Since extraction is performed, even in a table with a mixture of horizontal and vertical writing frames, the character lines within each frame can be extracted with high precision, and therefore the i8g! Accuracy can be increased.

[Brief explanation of drawings]

FIG. 1 is a block diagram showing an embodiment of the present invention, FIG. 2 is a flowchart of processing, and FIG. 3 is a diagram showing an example of a table in which horizontal writing and vertical writing are mixed. 101... Binary image input unit, 102... Binary image memory, 103... Table area recognition unit. 104...Table area image memory, 105...Main scanning direction line segment extraction unit, 106...Main scanning direction line segment coordinate memory, 107...
Sub-scanning direction line segment extraction unit, 108... Sub-scanning direction line segment coordinate memory, 109...
Frame recognition unit, 110...Frame coordinate memory, 1...Frame area extraction unit, 2...Frame area image memory, 3...Line direction determination unit, 4...Circumscribing rectangle extraction unit, 5. . . . Circumscribed rectangle coordinate memory, 6 . . . Line image extraction section, 7 . . . Line image memory, 118 . . . Recognition section. Figure 1

Claims (1)

[Claims] (1) In the table character recognition method, a frame surrounded by line segments in the main scanning direction and sub-scanning direction is extracted from a table area of a document image, and character lines within each frame are extracted and characters are recognized. It is determined whether the character line in each frame is written horizontally or vertically depending on the length of the frame in the main scanning direction and the length in the sub-scanning direction, and the character line in each frame is determined according to the result of this determination. A method for recognizing characters in tables characterized by switching extraction methods.