Automatic Video Text Detection and Localization Based on Coarseness Texture

Video text recognition is crucial to the research in all video indexing and summarization and has been used in video semantics analysis. Video text detection and localization is important for video text recognition. In this paper we present a new approach to implement video text detection and localization. In the text detection, we perform motion detection in 30 frames to get the motion mask. Then we get the synthesized frame based multi-frame integration. Finally, we retrieve an edge map of the synthesized frame using wavelet, and we detect text rows based on statistic coarseness of the edge map. In text localization, we locate the accurate boundary of text row based on edge map. Then we use the multi-frame verification (MFV) to remove the candidate text regions with motion occurrence on 30 consecutive frames. Experimental results show that our method is able to obtain a robust performance in the variation of language, font, size, color.