Research On Uyghur Text Extraction In Video Images

The rapid development of multimedia and network information technology, a large number of video images appear in the TV broadcast, digital libraries and on the Internet. The quantity of the information becoming bigger and bigger, the significance is also more and more important. So study for video image is a current hot topic. Xinjiang is in a special location with rich natural resources, half of the people are Uyghur. And the Uyghur fonts is very similar with Arabic, so the research of Uighur character recognition from video images is not only conducive to the improvement of the level of information processing in Minority, but also provided a reference for Arabic countries, what's more, further improve the level of exchange between Xinjiang and neighboring countries.Extracting the information from video images, first to find the location of the text information, extract the text area, and then input to the ORC system for recognition, thus finally completing the transfer of picture information. Extraction is the basis of recognition, only put forward a complete, clean text area, can come out better identify. The text image is divided into two categories:scene text and artificial text. The video images include these two types and maybe mix them in one video image. In the processing of multi-line text, firstly extract the edge of the image, to remove some of the noise by corrosion-expansion method, retaining the text area; Then set the conventional length, width, area ratio threshold to delete some parts of non-text area; then make angle correction for regions probably contain tilt text,compare the results for text and noise according to the set rules, removing more complex noise, to further determine the candidate region; Finally, we applied the polygon location algorithm to the text area, binarized the located text image to extract text.The extraction method is relatively new. It's comprehensive and available for single-line, multi-line, horizontal, tilted text. Polygon extraction is of great advantage compared to conventional quadrilateral, it is can reduce more noise besides the text, to ensure the accuracy of binarization and extraction. The experiments show that the proposed method applying well for Uyghur extraction, with high accuracy and being a higher level of research.