Research On Key Technology In Chinese Characters Recognition In Nature Scenes

Natural scene text contains rich information. Automatic access to text information of images can help people better understand the meaning of image and better process it. Such as storage, compression, retrieval. The key technology in Chinese characters recognition in nature scenes is researched in this paper. In the appendix, the establishment of on-line Chinese characters recognition database is introduced.The main contents of dissertation are as follows:1. It's a bottleneck impeding that there is no open, accurate and common database for the location and recognition algorithm study of Chinese natural scenes. So a database,which is named XD_Text L&R Database,used for text location and recognition embedded in nature scenes is established. The coordinates and meanings and some other information of the text region in the natural scene were demarcated.2. After a natural scene text localization algorithm, it will produce positioning inaccurate, false negatives, a number of characters connected, tilt and other issues. In response to this situation, a pretreatment algorithm of Chinese character recognition is proposed. Several methods are contained in this algorithm,such as binarization,character color extraction,color clustering,Character Segmentation. The rule of combining the text region is also proposed.3. A method for classifier integration is proposed. The euclidean distance classifier and the neural network classifier is used for classifier integration. The two classifiers are parallel. A vote law is proposed,and the integrated classifier could be adaptive when the vote law is put in use.4. An on-line handwritten Chinese character recognition database is built,which is named XD-MBOHD. The Chinese characters , numbers , letters ,common punctuation marks are all included in the database. The way how to collecting data information is also can be used in other domain,such as on-line signature recognition,handwritten identification.