Study Of Text Extraction Algorithm For Network Animation Filtering

At present, It improving people's life and communication style significantly by the great progress and combination among Computer, Multimedia and Communication technologies. On the Internet, more and more web pages with audios, videos, images and animations are emerged and became very popular. On the one hand, It can convey information between different people efficiently. On the other hand, many animations usually are advertisements and contain only junk information. How to filter these junk information in digital animations and give much more convenience for client to obtain valuable information has became a hot problem for researchers. Under this circumstance, this dissertation propose an animation filtering algorithm that base on text edge-like detection uses DCT method. Several work have been done to achieve this goal in the dissertation:1.Analyze the basic features of GIF animation which is the most popular animation on the Internet and provide a frame extraction paradigm of animation base on these features.2.Propose a DCT-based text edge detection algorithm base on basic animation features analysis and illustrate the it's principle and workflow. Besides, this dissertation introduce several basic digital image processing methods and their usage in this algorithm, includes Discrete Cosine Transform, Mathematical Morphology, Bilinear Interpolation and a few typical digital image processing methods. Finally, we can get a text edge-like image for single animation frame which is valuable for text localization and extraction.3.Propose a text localization algorithm which combine the multi-resolution technique and projection-based text localization. This algorithm can locate the text blocks'position in animation frame exactly. At the end of this algorithm, we segment these text blocks from animations and act them as input for OCR system.4.Introduce the state-of-art technologies for text detection and localization and relate to animation filtering use content-based method, such as OCR, Text Classification.The algorithm was implemented on Matlab platform, experimental results show that this approach is efficient to detect and locate text occurrences in network animations. Meanwhile, it is valuable for animation filtering and improves the filtering efficiency significantly.