KR19990016302A - How to recognize documents in chart form - Google Patents

Abstract

The present invention relates to a document recognition method in the form of a chart for recognizing a chart format consisting of vertical lines and horizontal lines in an OCR or a form document recognition system. There are a variety of problems such as inaccurate table image lines that are blurry, missing, or noisy. Accordingly, the present invention provides a first step of extracting a line component using an opening technique using a specific mask, a second step of extracting feature points having a direction component from the line component extracted in the first step, and the second step. The third step of reconstructing the chart using the feature points extracted in the step and the fourth step of recognizing the internal items by extracting the cells of the chart reconstructed in the third step are performed. It improves the recognition performance of documents and has a very good effect on document recognition such as automatic credit card recognition and bank slip automatic recognition.

Description

How to recognize documents in chart form

The present invention relates to a method for recognizing a chart and a form consisting of vertical lines and horizontal lines in an OCR or a form document recognition system. A method of document recognition in the form of a table.

1 is a flowchart illustrating a method of recognizing a document in a conventional chart form. As shown in FIG. 1, a first step of extracting lines by obtaining accumulated black pixels in a vertical direction and a horizontal direction with respect to an input binary image, and The second step of configuring the layout of the table using the lines extracted in the first step, and the third step of extracting and recognizing items in each cell after the configuration in the second step.

Looking at the prior art made of each step as follows.

With respect to the binary image input through the scanner, the black line is extracted by extracting the accumulated black pixels in the vertical and horizontal directions of the table.

In this way, a projection profile is obtained by extracting a line by obtaining a cumulative black pixel.

Doing this recursively extracts the vertical and horizontal lines.

The layout of the table is constructed using the lines extracted in the step S11. (S12)

As in step S12, the layout of the table is configured, and then the items inside each cell are extracted.

Recognize this extracted item.

Through the same process as above, it is possible to analyze and reconstruct the table by extracting and recognizing it.

However, the use of the projection line profile technique in the prior art as described above is time consuming, and lines of clean and inaccurate table images are blurry, missing, or noisy. There are various problems.

Accordingly, an object of the present invention for solving the conventional problems as described above is to provide a document recognition method in the form of a chart to significantly improve the recognition performance of a document containing a table such as OCR or slips.

Another object of the present invention relates to a method of recognizing a document in a tabular form to extract a line using an opening technique and extract a feature point having a direction component to accurately reconstruct the chart.

1 is a flowchart of a method of recognizing a document in a conventional chart form.

2 is a flowchart illustrating a method of recognizing a document in the form of a chart of the present invention.

FIG. 3 is an explanatory diagram showing a line component extraction state using an opening technique in FIG. 2; FIG.

4 is an explanatory diagram showing nine kinds of feature points in FIG.

FIG. 5 is a diagram illustrating types of feature points according to a combination of direction points in FIG. 4. FIG.

6 is an explanatory diagram showing a network composed of feature points extracted using FIGS. 4 and 5;

The present invention for achieving the above object is a first step of extracting a line component using an opening technique by a specific mask, and a second step of extracting a feature point having a direction component from the line component extracted in the first step And a third step of reconstructing the chart using the feature points extracted in the second step, and a fourth step of recognizing internal items by extracting the cells of the chart reconstructed in the third step. .

Hereinafter, the present invention will be described in detail with reference to the accompanying drawings.

FIG. 2 is a flowchart illustrating a method of recognizing a document in a chart form according to the present invention. As shown in FIG. 2, a first step of extracting a line component using an opening technique using a specific mask and a step extracted from the first step A second step of extracting a feature point having a direction component from a line component, a third step of reconstructing a chart using the feature points extracted in the second step, and extracting a cell of the chart reconstructed in the third step A fifth step is performed to perform the fourth step of recognizing the item.

Referring to the operation and effect of the present invention made of each step as described above in detail as follows.

When a table of binary images is input through a scanner, an opening technique is applied to the input binary images.

The opening technique refers to the application of the dilation technique after the erosion technique.

The erosion technique is a method of placing a specific mask on an image and marking the center point of the mask when the mask is included in the image. The dilation technique is a method of filling a specific mask with respect to each black pixel of the image.

Therefore, when the dilation technique is performed after the erasure technique using a long segment mask, only a part corresponding to the segment shape remains in the image and the rest is removed.

In addition, the middle part of the broken line in the original image is filled.

Therefore, by using a horizontal segment mask and a vertical segment mask, an image in which only horizontal and vertical components remain.

That is, as shown in FIG. 3, the line component using the opening technique is extracted.

The outline is extracted from the line component extracted in step S101.

4, the left and upper sides (hereinafter, abbreviated as upper left (a)) of the line as shown in FIG. 4, the left and lower sides (hereinafter, abbreviated as lower left (b)) as shown in FIG. 4, and FIG. Four types of corners are extracted: the right side and the upper side (hereinafter, abbreviated as upper right c) in FIG. (S102)

The combination of the four corners extracted as described above may determine the type of the feature point having the direction component as shown in FIG. 4.

That is, the feature points 1, 2, 3, and 4 are points formed by the corners themselves, and the feature points 5, 6, 7, and 8 are the tangent points and include two corners. .

As described above, the types of the feature points according to the combination of the corners are as shown in the table of FIG. 5.

For example, feature point 9 is the first consisting of upper left and lower right (a, d), the second consisting of lower left and upper right (b, c), and the third consisting of upper left, lower left and upper right ( a, b, c), the fourth consists of the lower left and the upper right and the lower right (b, c, d), the fifth the upper right and the lower right and the upper left (c, d, a), the sixth the lower right and The seventh case is composed of upper left and lower left (d, a, b), and the seventh case consists of upper left and lower left and upper right and lower right (a, b, c, d).

Therefore, a network is constructed using the feature points having the direction component obtained in step S102.

Each feature point has its own location and information of neighboring feature points such as left, right, up and down.

Therefore, the feature points missing from the original image can be restored by making the most of the network information.

The finished net is the perfect layout of the table.

You can use this layout to reorganize the table at any time and into the exact shape and location you want.

FIG. 6 shows a network of the table constituting FIG. 3.

In this way, when the layout of the table is finished, each item necessary for each cell is extracted by using a projection profile or a connected component technique.

When the extraction of each item is finished, the extracted item is transferred to the recognizer and recognized (S104).

As described above, a line component is extracted, and feature points are extracted from the extracted line components to form a table layout, and necessary items for each cell are extracted and sent to a recognizer.

As described above, the present invention improves the analysis and recognition performance of the tabular document by configuring the layout of the table using the feature points of the line component, and also has a very good effect on the document recognition such as automatic credit card recognition and automatic bank slip recognition. There is.

Claims (2)

A first step of extracting a line component using an opening technique by a specific mask, a second step of extracting a feature point having a direction component from the line component extracted in the first step, and a feature point extracted in the second step And a fourth step of reconstructing the table using the second step, and a fourth step of recognizing the internal items by extracting the cells of the table reconstructed in the third step. The method of claim 1, wherein the feature point having a direction component is extracted using four types of corners: left and top, left and bottom, right and top, and right and bottom of the line.