KR910009929B1 - Method for extracting document region for document recognition - Google Patents

Abstract

No content.

Description

Document Area Extraction Method for Document Recognition

1 is an exemplary view showing the scanning direction and block extraction data of the document of the present invention.

2 is an exemplary diagram of block extraction according to FIG.

3 is an exemplary view of block extraction in the scan lines B and C directions according to FIG.

4 is a signal flow diagram of the present invention.

* Explanation of symbols for the main parts of the drawings

BP1-BP3: Current pixel value in B direction scan

CP1-CP3: Current pixel value in C direction scanning

Cont1: Left contour of block Cont2: Right contour of block

P: Left pixel value of current pixel

The present invention relates to the separate extraction of individual regions within a document for recognition of document units, and particularly, to extract document regions for document recognition so that the same regions are not duplicated and extracted by not repeatedly tracking the contours of each blocked region. It is about a method.

In general, a contour following method is widely known as a method of extracting an area by tracing a contour of a block, but this method requires a precondition that there is no hole in the block.

If there is a hole in the target block or another nested block in the block, and a plurality of such blocks exist in one image, and another block overlaps the distorted block, etc. You can't simply apply this method directly. Therefore, a considerable problem has arisen to extract a block using the conventional contour tracking method in an image such as a document having a plurality of blocks of various forms.

The present invention prevents the internal contour extraction of the block and the repeated extraction of the same block when extracting only the outer contour of the block using the contour tracking method from the document image in which various types of blocks exist in view of such a conventional problem. In order to enable contour extraction by the contour tracking method, the present invention will be described in detail below with reference to the accompanying drawings.

The document is composed of large text (Text, text) area, graphics and graphic areas such as pictures and charts, and image area, and document recognition separates and extracts each area of the document. This means that proper processing according to the characteristics of each area is performed.

Therefore, the first step for document recognition is to block each area of the document and extract these blocks.

In order to extract the area of the document, the outer contour information of the block is required, and therefore, the inner contour of the block is unnecessarily tracked (that is, when a hole exists inside the block) or the same block must not be duplicated and extracted. To this end, the present invention detects contour pixels of blocks already extracted while scanning an image horizontally, and performs raster scan of an image using contour information logically configured for the number of repetitions. In this case, two pixels are detected while passing through the extracted blocks.

However, when multiple contour pixels are detected in succession, they assume that one contour pixel has been detected, and contour tracking indicates whether to start block extraction from this pixel only when the pixel value of the examined pixel is "1" block. In this case, the contour information extracted so far becomes the basis. At this time, the contour information to be extracted, that is, the left contour of the contour, the right contour of the block, and the values of the immediately left pixels X-1 and Y of the current pixel coordinates X and Y are used as the reference. At this time, the contour information to be extracted is cleared after a horizontal scanning, and then it is determined whether to extract a block by tracking the contour from the current pixel using the current pixel value and the contour information in the state.

Referring to the accompanying drawings, the operation and effects of the present invention are as follows.

4 is a signal flow diagram of the present invention. As shown in FIG. 4, the coordinate (X = 1, Y = 1) value of the pixel currently being investigated is "1" and the present pixel value BP1-BP3 (CP1-CP3). Only when it is "1", it is decided whether or not to extract the trace block starting from this pixel.

When the current pixel values BP1-BP3 (CP1-CP3) are "1", the left contour (Cont1) of any block already extracted while scanning to this point has not been detected, that is, Cont1 = 0 and immediately left Only when the pixel coordinates (X-1, Y) are not outlines, i.e., when the left pixel value P of the current pixel is zero, the block is set with the current pixels BP1-BP3 (CP1-CP3) as a starting point. Extract. Herein, the pixel value P of the current pixel represents the left pixel value P. If the pixel value is the left outline Cont1, the pixel value P of the current pixel has a value of "1". The immediately left pixel value P has a zero value.

In Fig. 4, all pixels designated as outlines during block extraction change their values from “1” to “2”, and the extracted data from the block is the coordinates (X) (Y) of the starting pixel and the vertical and horizontal straight lines of the block. The area of the block is determined by obtaining the length.

When the block is extracted, the values of the left outline Cont1 and the right outline Cont2 of the block are detected as “ø”. That is, the information detected so far is the value of all pixel values Since it was used as a criterion of block extraction in “2”, it resets all and extracts contour information again to extract a new block, and then the current pixel values (BP1-BP3) (CP1-CP3) become “2”. Since the left pixel value P of the current pixel is set to " ø "

If the current pixel values BP1-BP3 and CP1-CP3 are the contour pixel values CP = 2, the values of the contour information are adjusted by referring to the contour information extracted so far.

If no left outline Cont1 of any block is detected and the left pixel value P of the current pixel is not a contour pixel, that is, P = 0, then Cont = 1 of the left outline (already extracted) of any block Is considered detected.

Here, the left outline Cont1 of the block and the right outline Cont2 of the block are set to "1" when the contour pixel is detected and set to "ø" if not detected. In addition, the right contour Cont2 value of the block is not compared because the right contour is always detected only after the left contour is detected. Therefore, the value of the right outline Cont2 of the block is designated only by the left outline Cont1 information of the block.

If the left outline Cont1 of the block is detected and the left pixel value P of the current pixel is not a contour pixel, the left pixel value P becomes “ø”, and the current pixel is designated as the right outline pixel. do. In addition, when the pixel value is a background value other than “1” or “2”, it is necessary to reset the left outline Cont1 and the right outline Cont2 of the block to “ø”, which are all contour information. The coordinate (X) (Y) designation operation of the next pixel is performed.

That is, if the right outline Cont2 of the block is detected or the left outline Cont1 of the block is detected and the left pixel is the outline pixel, all the contour information is reset to “ø” and the last pixel of the horizontal line According to the determination of the coordinate (Y) of the last horizontal line, the left outline (Cont1) of the block and the right outline (Cont2) of the block, the left pixel value (P) and the coordinate (X) of the current pixel are “ø” and the coordinate (Y) is Y + 1 and adjusts the coordinates at the next pixel (X = X + 1).

After the irradiation to the last pixel of the horizontal line, the coordinate “1” of the Y axis is increased and the left outline Cont1 of the block, which is all the contour information, and the right outline Cont2 of the block are reset. Thus, when the investigation is completed to the end of the image, the algorithm is finished.

FIG. 2 is an exemplary diagram of block extraction of FIG. 1 and supplemented with FIG. 3 as follows.

FIG. 2 scans in the A direction, starting from the pixel AP in the A direction, and the value of the pixels corresponding to the contour of the extracted block is changed from “1” to “2” as a result of the block extraction.

In this situation, the block extraction determination process in the B direction scan and the C direction scan is shown in FIG. 3A.

That is, in the scanning in the B direction, the current pixel values BP1, BP2, and BP3, which are all three "1", are detected, but the left outline Cont1 of the block, which is the contour information at that time, and the right outline Cont2 of the block. The current pixel values BP1 and BP2 do not satisfy the condition that the left outline Cont1 of the block is “ø” and the left pixel value P of the current pixel is “ø”. Block extraction does not start from the pixel values BP1 and BP2. However, in the case of the current pixel value BP3, block extraction is not attempted at the pixels CP1, CP2, and CP3 by the right outline Cont2 of the block under the condition of the contour information.

As described above, the algorithm of the present invention is characterized in that by determining whether to extract a block based on the contour information, the outer outline is obtained so that the blocks of the entire document do not overlap and the region is separated and extracted as the coordinates of each block.

Claims (3)

In the method of extracting a region of a document by contour tracking, the left and right contour lines (Cont1) and (Cont2) pixels of already extracted blocks are detected and detected to prevent overlapping of individual regions while scanning an image horizontally. The left and right contours Cont1 and Cont2 are detected to determine whether to extract the individual areas by tracking the contours from the current pixel using the combination of information, and the left contour Cont1 of the block and the right contour of the block ( The Cont2) value is reset, and if the contour pixel of the contour information is detected, it is set to "1", and if it is not detected, it is set to "ø", the left outline Cont1 of the block is "ø", and the left pixel is the contour pixel ( If not P, the left outline Cont1 is assumed to be detected, and the left and right outlines Cont1 and Cont2 of the block are set to "ø" if the outline pixel is not detected as "1", and the left and right outlines areThe document characterized in that the document area is extracted by specifying the current pixel values BP1-BP3 and CP1-CP3 as the right outline Cont2 pixels if the left pixel is not the outline pixel P (ø). Document Area Extraction Method for Recognition. 2. The pixel according to claim 1, wherein the background pixel (background valud, θ) is detected, the left and right contours (Cont1) (Cont2) have already been detected, and the left and right contours (Cont1) (Cont2) are reset to "ø" to A method for extracting a document area for document recognition, characterized by extracting an area of a document by displaying whether the left and right contours (Cont1) and (Cont2) of the block can be detected. The method according to claim 1, wherein if the current pixel is detected with the information value 1 of the block, and both the left and right contours Cont1 (Cont2) are "ø", the contour is traced from the position of the current pixel, and an individual area is extracted to extract the document. Document area extraction method for document recognition, characterized in that for extracting the area.

Images