The Research Of Video Text Extraction Under Complex Background

With the rapid development of communication technology, computer technology,broadband network technology and information technology, how to quickly find the contentof interest for the user in the massive video data, has become the urgent problem ofinformation society, the development of content-based video retrieval system has become avery practical value and wide prospect of the subject in the field of multimedia informationprocessing at present. The text of Video contains rich semantic information, therefore, videotext extraction plays an important role for the development of video retrieval technology.The extraction of video text generally includes four processes: text detection, textlocalization, text segmentation, text recognition. Based on analysis of the common textpositioning, text segmentation and text recognition and compare the advantages anddisadvantages, summarizes up the proposed a coarse positioning video text diction based onthe edge and the algorithm of wavelet transform, then use the support vector machine (SVM)for precise positioning. The method uses the strong edge information of text and the strongcontrast feature of background for text detection, to eliminate unnecessary image area andretain image area, has a higher precision. Then this paper proposes segmentation algorithm isa local threshold value method combined with FCM，effectively inhibit the effects of noise onthe segmentation results. Finally, using the identification method based on sample pictures,first of all training samples, and then use artificial neural network to recognize words.Through the use of the above algorithms, the paper completed the construction of the videotext extraction system. The simulation experimental results show that the improved algorithmand the effectiveness of the proposed text extraction system.