Research Of Skew Correction Algorithm For The Complex Document Images

With the rapid development of information technology, coupled with the enhancement of people's awareness of environmental protection, each industry promote the paperless office.through the network, the electronic documents can be easily circulated among the various departments, It greatly reduces the cost of manpower and materials. So the electronic document is slowly replacing the paper documents.Electronic document is constantly changing people work patterns and lifestyles, such as electronic archives, bank notes, online checking and so on. Thus it seems more and more important and can't wait to turn the paper documentation into electronic documents rapidly and correctly.When converting paper documents into electronic documents, it is inevitable to produce a certain degree of tilt error because of factitiousness or the equipment. The tilted images not only affects visual effects, and has important effects on the version analysis, therefore the research of image tilt algorithm seem to be very necessary.This paper studies the complex document images skew correction, that is, the document contains not only text but also images of documents.On the basis of the existing text skew correction, put forward two different kinds of graphic image skew correction algorithm. The first method is through a series of image preprocessing,such as gray processing, binarization and segmentation; then combining Radon Transform and data analysis method to exclude blocks which interfere the detection of sub-blocks; Last merge these blocks, and again using Radon Transform to detect angle and correct the image. The second method is based on the algorithm of calculating the sunspot cycle, using the text of periodic to get more precise angle. The first step is also through a series of image preprocessing,such as gray processing, binarization and segmentation; The second step is using Radon Transform get each angle of blocks, on inclination data for Fast Fourier Transform to get histogram of frequency domain and the relation diagram between power and cycle of each sub block. The last is according to text of periodic and size of power to filter the blocks, thus it will get the blocks which contain many text, and then usingRadon Transform to detect angle and correct the image. Using either of the above algorithms may be get the ideal angle.According to the analysis results show that compared with the related algorithms, these two algorithms are applicable-180-180 angle detection and correction, compared to the previous algorithm has been greatly improved in the range of detection angle, and for detection of text-images, with high accuracy and robust advantage. The first method has a high degree of efficiency, but there is some error. The accuracy of the second methods is relatively high, and it has a more accurate angle measurement for the complex structure and the content of the digital image, thus greatly improving the scope of application of the algorithm.