Text Recognition And Post-processing In Natural Scene Based On The Handheld Terminal

With the rapid development of the information society, more and more people beginto pay attention to multimedia technology. Handheld terminal products with a built-incamera have been in widespread use, and smart phone is the most typical representative.As smart phone becomes popular, it not only makes people more easily to get naturalscene images in everyday life, but also brings broad application prospects to textinformation extraction in natural scene images.Because text in natural environment has various colors, Chinese characters changerange in size and have different arrangement, image background is relatively complex,text location and recognition in natural scene is a very challenging task.Based on handheld terminal, this paper has done a research on text location、recognition and post-processing in natural scene images, and proposed a complete textextraction system based on handheld terminal. The main content is as follows:1. Text Location in Natural Scene ImagesThis paper puts forward a method of semi-automatic text location based on colorclustering. Manually select a certain target color point through the handwritten pen ofhandheld device, then analyse the image and make color clustered, finally do connecteddomain processing and give the results of text location.2. Text Recognition in Natural Scene ImagesGet the position of the text through text location to natural scene image, then dobinary processing to the text area of image and remove isolated noise points, furtheroperations are single character cutting, clipping and normalization, and then extract thefeature of line direction to normalized image, at last put the feature vector into theEuropean distance classifier and get the results.3. Post-processing of text Recognition in Natural Scene ImagesPost-processing to the results of recognition is needed after text recognition, whosepurpose is to improve the text recognition rate. Firstly, establish associated thesauruswhich is necessary to phrase association, and then design the storage method ofvocabulary entry based on special marks; secondly, sort the rectangular frames to ensurethat there is possibility to make phrase between two adjacent Chinese characters; finally,give the specific implementation of forward-backward search and adaptive phraseassociation procedure.