JPH03172984A - Table processing method - Google Patents

Abstract

PURPOSE:To accurately segment a character in a frame even when an image is inclined by removing a ruled line for a remarked frame or a black-connected circumscribed rectangle contacting with an adjacent frame in the frame. CONSTITUTION:A main scanning direction segment extracting part 15 traces black elements connected in the main scanning direction in a table area image, extracts a rectangle surrounding a ruled line in the main scanning direction and stores the start point coordinates (Xs, Ys) and end point coordinates (Xe, Ye) of the extracted rectangle in a main scanning direction segment coordinate memory 16. A subscanning direction segment extracting part 17 similarly stores the start and end point coordinates in the subscanning direction in a subscanning direction segment coordinate memory 18. A black-connected circumscribed rectangle extracting part 23 extracts a black-connected circumscribed rectangle from an intra-frame image and stores the coordinates of the extracted contents in a circumscribed rectangle memory 24. A line segmenting part 25 regards the black-connected circumscribed rectangle in contact with the frame as a part of a ruled line constituting the frame and removes it. A character segmenting/recognizing part 27 segments a character from a character line image and recognizes the segmented character.

Description

DETAILED DESCRIPTION OF THE INVENTION [Field of Industrial Application] The present invention relates to a method for processing a table in an image of a single document, form, etc. in a single character recognition device or the like.

(Conventional technology) When processing document images in a character recognition device.

An image is often divided into a text area, an image area such as a photograph or diagram, one table area, etc., and different processing is performed on each area.

For table areas, use IR coordinates for ruled lines.

The method used is to recognize each frame in a table, cut out characters from the image within each frame, and recognize them.

Conventionally, in processing such tables, the rectangles of the ruled lines in the main scanning direction and sub-scanning direction that make up the frame are extracted, and the coordinates inside this rectangle (viewed from inside the frame) are used to define the frame. ing.

[Problem to be solved by the invention]

However, there is a problem in that when a document image is input at an angle, the characters within the frame cannot be correctly extracted.

For example, in the case of an image as shown in FIG. 4, solid lines are ruled lines (line segments), but they are extracted as rectangles like broken lines. Figure 5 shows the main scanning direction (X direction)
is an explanatory diagram of the rectangle of the ruled line, 51 is the actual inputted ruled line, and 52 is its rectangle. Then, when recognizing a frame surrounded by ruled lines, conventionally the coordinates inside the rectangle of the ruled lines are used. For example, in the case of a rectangle in the main scanning direction that forms the upper side of the frame, the coordinates Ye Define the Y coordinate of the top side of the frame. Therefore, in the case of the tilted image shown in FIG. 4, the hatched area will be extracted as the frame area, making the frame narrower than it actually is. As a result, characters that are close to the ruled lines within the frame may protrude from the frame.
It may not be possible to cut it out properly.

An object of the present invention is to provide a table processing method that allows characters in a table to be correctly extracted and recognized even when a document image is tilted manually.

[Means to solve the problem]

The table processing method of the present invention extracts a rectangle that includes ruled lines in the main scanning direction and layer scanning direction in a table area, and creates a frame surrounded by the ruled lines using coordinates outside the ruled line rectangle (viewed from inside the frame). This method is characterized by recognizing the black connected circumscribing rectangles within the frame, removing the circumscribing rectangles touching the frame, and using the remaining circumscribing rectangles to cut out the characters within the frame.

[Operation] Since the frame is recognized using the coordinates outside the ruled rectangle, even if the document is tilted, the width of the recognized frame will not be extremely narrower than the actual width, so there will be no missing characters. It can be prevented.

For example, in the case of an image of a tilted table similar to the image shown in FIG. 4, according to the present invention, the recognized frame becomes wider as shown in FIG.

On the other hand, to use coordinates outside the ruled rectangle.

A portion of the ruled lines forming the frame is included inside the recognized frame. However, such line segment portions of ruled lines can be removed by excluding those that touch the frame among the black connected circumscribed rectangles within the frame.

Therefore, even in the case of a tilted document image.

It becomes possible to correctly extract and recognize the characters in the table.

〔Example〕

FIG. 1 is a block diagram showing an embodiment of the present invention, and FIG. 2 is a flowchart of processing.

The document is read by the binary image input unit 11 such as a scanner, and the binary image is stored in the binary image memory 12 (processing step 31). The image is recognized and stored in the table area image memory 14 (processing step 32).This table area recognition can be performed by automatically recognizing the area using run length distribution or by using a mouse or the like from outside. By specifying H! There is a way to do &i.

The main scanning direction line segment extraction unit 15 traces the black pixels connected in the main scanning direction with respect to the table area image, extracts a rectangle surrounding the ruled line in the main scanning direction, and extracts a rectangle surrounding the ruled line in the main scanning direction, and its starting point coordinates (Xs,
Ys) and end point coordinates (Xe.

Ye) is stored in the main scanning direction line segment coordinate memory 16 (see FIG. 5, processing step 33).

The sub-scanning direction line segment extraction unit 17 similarly extracts the rectangles of the ruled lines in the sub-scanning direction from the table area image, and stores the coordinates of the start and end points in the sub-scanning direction line segment coordinate memory 18 (
Process step 34).

Next, the frame recognition unit 19 recognizes a frame based on a combination of the ruled lines in the main scanning direction and the ruled lines in the sub-scanning direction, and stores the coordinates of, for example, the diagonal vertices of this frame in the frame coordinate memory 20 (step 35). At this time, as mentioned above, the coordinates of the outside of the rectangle of the ruled lines that make up the frame (the coordinates of the top side for the top ruled line, the bottom side for the bottom ruled line, the left side for the left side ruled line, and the bottom coordinate for the right side ruled line) to find the coordinates of the diagonal vertices of the frame. In this way, the frame area shown as the hatched area in FIG. 3 is recognized.

Next, the frame area extraction unit 21 cuts out an image within the frame from the table area image according to the frame coordinates, and stores it in the frame image memory 22 (processing step 36).

The black connected circumscribed rectangle extraction unit 23 extracts a black connected circumscribed rectangle from the frame image and stores its coordinates in the circumscribed rectangle memory 24 (processing step 37).

The line cutting unit 25 checks the contact between the circumscribed rectangle and the frame by comparing the coordinates of the extracted black-connected circumscribed rectangle with the coordinates of the frame, and extracts the black-connected circumscribed rectangle that is in contact with the frame with the border of the ruled line constituting the frame. It is considered to be a part and removed (processing step 38).
Then, for the remaining black connected circumscribed rectangles in the frame, the character size is estimated and integrated to generate a character line within the frame (integrated rectangle of the circumscribed rectangles that make up the grapheme), and any necessary corrections are made. Alternatively, the image of the final character line is cut out from the frame image memory 22 and stored in the line image memory 26 (processing steps 39, 40, 41, 4).
2).

Next, the character cutout/recognition unit 27 cuts out and recognizes characters from the one character line image (processing step 44).

〔Effect of the invention〕

As is clear from the above description, according to the present invention.

Even if the document image is tilted manually, the recognized frame will not become inappropriately narrower than it actually is.
It is possible to prevent missing character images, and it is also possible to remove the ruled lines that make up the frame before cutting out characters by removing the rectangles that touch the frame by connecting black in the recognized frame. .

It is possible to accurately cut out and recognize characters.

[Brief explanation of the drawing]

FIG. 1 is a block diagram showing one embodiment of the present invention. Fig. 2 is a flowchart of the process, Fig. 3 is an explanatory diagram of frame recognition, Fig. 4 is an explanatory diagram of problems in frame recognition by the conventional method,
FIG. 5 is an explanatory diagram of a ruled line rectangle. 11... Binary image input unit, 13... Table area recognition unit. 15...Main scanning direction line segment extraction unit. 17... Sub-scanning direction line segment extraction unit. 19... Frame recognition unit, 21... Frame area extraction unit. 23...Black connected circumscribed rectangle extraction part. 25...Line cutting section, 27...Character cutting/recognition section. Part 2 Figure 3 Ironworks with wealth n a certain giant #1 job old shoulder 1 tanro ===Ko style territory・Figure 5→-×

Claims (1)

[Claims] (1) In the table area, extract the ruled rectangles in the main scanning direction and the sub-scanning direction, recognize the frame surrounded by the ruled lines using the coordinates outside the ruled line rectangle, and calculate the black connected circumscribed rectangle within the frame. A table processing method characterized by removing a circumscribed rectangle touching the frame, and using the remaining circumscribed rectangle to cut out characters within the frame.