JPH083832B2 - Document image structure extraction method - Google Patents

Description

TECHNICAL FIELD The present invention relates to a method for recognizing and extracting a region such as a character line, a graphic portion, and a sentence of a document image.

(Prior Art) Conventionally, as a structure extraction method of a document image, a method of cutting out a continuous region of non-white pixels in units of black pixels or coarse meshes, approximating this to a rectangle, and connecting adjacent rectangles is often used. Known (see, for example, Computer Vision 22-1 1983.1.27).

However, with this method, all sentences and figures written in the rectangular frame are hidden inside the frame. Also,
If there is a small black stain between the two areas, the two areas will be connected. For example, as shown in FIG. 2, if there are two pictures or diagrams shown by solid lines and there is a dirty part such as B or one thin line between the connected image areas A and C shown by broken lines, A -B-C are merged into one graphic area.

(Object of the Invention) In a document image, not only printed or filled black information but also white information (including margins) often represents a structure. The present invention provides a method for extracting a structure of a document image, which positively utilizes this white portion to extract an image area such as a character line, a graphic portion, and a sentence.

(Structure of the invention) Remove vertical and horizontal line segments from the target document image,
Next, the white part is divided into rectangles, and the resulting non-white part is connected and labeled to extract the text or graphic part.

(Embodiment) FIG. 1 shows processing steps of an embodiment of the present invention. First, in step 1, long line segment elements (only vertical and horizontal components) of the target document image are extracted and deleted. (For the method of extracting line sentences, see, for example, IPSJ 25th = 2nd half of 1982 = National Convention Proceedings 5B-4)

Next, in step 2, the image is divided into meshes.
A well-known method is the Quad-Tree method.
As shown in the figure, degenerate representations are sequentially made as 0 layer (original image), 1 layer (4 pixels of the original image are replaced by 1 pixel), 2 layers (4 pixels of 1 layer are replaced by 1 pixel) ,. There are various methods for this expression, but here, the sum T of four pixels (a, b, c, d), that is, T =
When a + b + c + d is above a certain threshold value, it is set to black, and otherwise it is set to white (0).

In step 3, the white area is divided into rectangles. The division method is
Reference Pattern Recognition Vol.11.pp297〜312 Aoki “RE
CTANGULAR REGION CODING FOR IMAGE DATA COMPRESSIO
The N "method may be used. Fig. 3 shows an example of the division.

Next, as step 4, labeling processing is performed on the non-white rectangular area in FIG. 3 (b). As a result, the character line and the graphic part are cut out into a rectangle. At this time, the area having a large vertical and horizontal width can be regarded as a figure, and the other areas can be regarded as a character or a text area. That is: ◎ The non-white areas that are long horizontally are highly likely to be character lines.

◎ Large non-white areas are highly likely to be graphs or pictures.

Finally, in step 5, the non-white areas are integrated and the image area of the document image is extracted.

This method is more effective in the following cases. A narrow non-white area having wide white areas on both sides like the area B in FIG. 2 is highly likely to be unnecessary noise. Therefore, the non-white area like D in FIG. 5 is erased,
The two white areas E and F can be integrated.

(Effects of the Invention) As described above, according to the present invention, a non-white area can be cut out as a large block, so that the structure of a document image can be extracted quickly and stably.

[Brief description of drawings]

FIG. 1 is a diagram showing processing steps of an embodiment of the present invention,
FIG. 2 is a diagram showing an example of a document image having two pictures or diagrams, FIG. 3 is a diagram showing an example of rectangular division of a white area, and FIG. 4 is a diagram showing an example of an image mesh division method. , Fig. 5 shows
It is a figure which shows the processing method when there exists a non-white area | region between two adjacent white areas. A, C ... Connected image area, B, D ... Narrow non-white area, E, F ... White area.

Claims (2)

[Claims] 1. A process for removing vertical line segments and horizontal line segments from a target document image, a process for dividing a white part into rectangles, and a process for performing a concatenative labeling process on a non-white part. A method for extracting a structure of a document image, wherein the extracted portion is extracted as a text or a graphic portion. 2. The process of dividing the white portion into rectangles, wherein when the plurality of white regions are adjacent to each other with a narrow gap, the plurality of white regions are regarded as one region. A method for extracting a structure of a document image according to the item (1).