The Implementation of Rolling Video Text Localization Based on FSVM

Video text contains abundant high-level semantic information, which is important to video analysis, indexing and retrieval. In this paper, fuzzy support vector machine (FSVM) is applied to distinguish background and text in a video sequence. Firstly, the video frame is divided into 8×8 blocks, and we extract the gray, edge and texture feature information as the training samples. Then FSVM is used to classify the samples and to get the candidate text regions. Lastly, according to the character of text region, the method is adopted to process the candidate text regions for getting the real text regions and to mark the text with text box so that the video localization can be achieved. The final experimental results show that FSVM is effective on rolling video text localization, especially on keeping the accuracy of the text localization with the complex background.