Research On Artificial Text Detection Methods In Video Based On Wavelet Transform

With the number of digital multimedia libraries increasing rapidly, the requirements for effective retrieval of, browsing and testing multimedia information have increased. Among those requirements, the text in images provides important semantic information, and is also relatively easy to be extracted than voice signal. Thus, the effective detection and extraction of text information is of great significance for video and image content analysis, understanding and content-based retrieval.The wavelet transform is a signal analysis method of the time—frequency. The ability of local features of signals in both time domain and frequency domain can make wavelet extract the characteristics of local information from image effectively .So the texture of text region can be accurately captured.As wavelet transform is applied to the text detection, the extraction of texture feature in the wavelet domain and the text detection methods based on wavelet are the main content in this thesis. First, the basic theory is introduced related to the subject: the basic concepts of wavelet transform and Multi-Resolution Analysis. Then, after the characteristics of each sub-band wavelet coefficients being analyzed, the improved feature of average texture energy is proposed. On the basis of the new feature, text detection and localization are implemented by clustering algorithm based on the new feature vectors and post-processing. And the new feature is compared by experiment with other feature. Finally, In view of the features ahead and experimental results, a new text detection method is proposed based on the texture total energy of wavelet. At last, Comparison is made by experiment with the above method.