Document image geometry fault correction method

Abstract

The invention relates to a method for adjusting the distorted image with complex text content in the field of the image processing technology of computer information processing. The current technology uses level projection to find the position of the text line, and impacts the text line center line to adjust it. But when the content occurs to the non-text area such as image, form and so on or the text bending strongly, it can't find the text line to do subsequent processing.

Description

Technical field [0001] The invention relates to image processing technology in the field of computer information processing, in particular to a method for correcting geometric distortion of document images. Background technique [0002] Distorted image correction is a very useful image processing technology. Generally, there are two types of correction methods. One is to set some known reference points on the image and perform correction based on the comparison of the reference points before and after the distortion; the other is It is no-reference correction, which is completely corrected by analyzing the characteristics of the image itself. [0003] For the former type, the general approach is to add some known reference points to the image through a certain method, that is, the correspondence between some pixels of the undistorted image and the coordinates of the corresponding pixels of the distorted image, as the basis for correction. For example, a layer of grid is pasted on...

AI Technical Summary

Problems solved by technology

If these features are combined with the complex factors of the layout, the method for analyzing text lines in the prior art will be completely ineffective: 1) when there are serious text bending, image, table interference, and text lines not running through the left and right, etc., through the horizontal Where the projected unobtainable text is located

2) After obtaining the approximate position of the text line, it is difficult to distinguish the text line from the image, table, lace, etc. If the image, table, and lace are regarded as the text line, due to their large differences in height and position, use the calculated The curve obtained by fitting the center line cannot reflect the real distortion trend

3) If non-text areas such as images, tables, and lace are not processed, the position of the corrected text line will deviate, while the position of the unprocessed area remains unchanged, so the relative relationship between the corrected and uncorrected areas will be different. Changes will occur, and in serious cases, the corrected area will fall into the uncorrected area, resulting in overlapping phenomena, causing errors in subsequent layout analysis, segmentation, etc., or even impossible

[0010] It can be seen that the prior art has only dealt with neat, slightly curved plain text document images

However, it cannot handle document images with severe curvature and relatively complex layout.

Images