Research On Embed Text Extraction From Still Images

Text in images and videos carries plenty of semantic information useful for understanding the content of images and videos.Thus, it makes text recognition very significant for understanding and retrieval of images and video.However, text is usually embedded in complex background of images, which makes direct optical character recognition almost impossible.Therefore, it becomes necessary to extract text from complex background before recognition.In the last decades, many efforts have been devoted to developing effective algorithms to extract text from complex background in images and video.However, the state-of-the-art of text extraction is far from perfect due to the great difficulty in discriminating text from complex background completely.In this thesis, my research work aims to improve the performance of text extraction from two aspects: more robust text segmentation algorithm utilizing hybrid information and effective post-processing techniques to eliminate background residues.The features extracted from strokes of characters are experimentally checked to evaluate their feasibility in discriminating text region from complex background regions,The main contributions of the thesis include:Based on the analysis of constitute of character, designed a method with constructing and describing the texture characteristics of text strokes by using statistical texture analysis . By sliding window to scan the text block area, extract the feature vectors for each window, and then apply the method of SVM to distinguish between two modes of text and non-text classes. This method has the less sensitive on color, font and style of text, which extend the general scope of the text detection.In the process of text segmentation, combined with the characteristics of the text characters in the image, proposed a targeted area histogram equalization method, in order to achieve the enhancement of the text area.A global threshold and the local threshold method of combining binary text images proposed, which made up the simple use of an effective global threshold or the problems of the local threshold .Design a multi-projection method to detect the direction of the tilt angle of the text, which could meet the requirement of the commercial OCR text recognition system in angle error of text .