Images and videos on the web and database are increasing rapidly. Analysis of such huge database can be done by using its content. The studies have proved that text in image or video acquires human attention dominantly. Hence text detection plays very important role in the information retrieval. Text detection end extraction plays very important role in information retrieval. Text detection is a challenging task due to complex background, font of the text, orientation of the text, contrast etc. In this paper, we are presenting a new approach to detect the text in complex images and videos. Multi-oriented text detection in video frames is not as easy as detection of caption or graphics or overlaid text which usually appears in horizontal direction and has high contrast compared to its background. Multi-oriented text generally refers to scene text which makes text detection more challenging and interesting due to unfavorable characteristics of scene text. Therefore, conventional text detection methods may not give good result for multi-oriented scene text detection. In the existing system, since video suffers from complex background and low resolution, it is hard to preserve the shape of every character. Therefore, the existing document analysis based methods may not be suitable for video text recognition. In this project the system proposed a new system to recognize video texts through binarization by introducing a Bayesian classifier. This shows that conventional binarization methods work well for graphics text of high contrast with plain background, but not for scene text because it suffer from the effects of illumination variations that are unpredictable. The method performs a histogram analysis for each sliding window over each linearly combined image. The height and width of the window are the same as the height of the word.