Research And Application Of The Tilt Correction Of Document Images Based On Recognition Feedback

You can output the contents of paper documents to editable computer text through OCR (Optical Character Recognition) technology. Currently, OCR technology is relatively mature, and it can get a high recognition rate for good quality scanned or shoot images. However, scanned paper documents or pictures will produce a phenomenon of document placed tilt inevitably, when we do the OCR recognition to the image which is collected under this conditions, it will produce a big error so that the recognition rate dropped sharply. Therefore, the correction of the skewed document images is necessary.When we correct the skewed document images, we will find that there are various factors can affect the results of correction, such as text fonts, fonts, scripts, and if it contains images, which will be affected the effect of correction algorithm to varying degrees, and also will bring a certain complexity to the correction of the skewed document images.The source of the subject is the national "Eleventh Five-Year" National Key Project Support Platform-"The Development of Reading Aids for the Visually Impaired".The main objective of the project is to develop a portable reading aids which is based on text recognition for the visually impaired person. The Reading aids make the printed text and electronic documents convert to voice and output it through the use of OCR technology and speech synthesis technology, so that the visually impaired can make the same as ordinary people to read newspapers and books. Before the OCR recognition, we need do some preprocessing for the document images, the correction for the skewed document images is a very critical step specially. It requires that the algorithm can get a better treatment effect for most of all the skewed document images that the algorithm should have a good versatility and we should measure the time complexity.For these purposes, I have proposed a new improved algorithm of angle detection for skewed document images after I have learned and researched variety classical algorithms of skewed document images. I have summarized and analyzed the strengths and weaknesses of these algorithms. And then I have combined with the specific requirements of the development of the Reading Aids for the Visually Impaired. The algorithm uses the detection method which is based on the feedback information OCR recognition. It can select out which candidate algorithm can correct the skewed document images better. So that the algorithm will overcome the flaw of one algorithm can not have good general condition of defects for various document types to some extent.