The Research Of Video Text Text Ddetection And Its Iimplementation

With the network technology and digital development, digital images and video grew very fast. At the same time, with the popularity of mobile digital camera device, users can easily take a variety of natural scenes in the digital image. To facilitate browsing, management and understanding of the images and video's content, a number of actual needs of digital image processing and understanding are generated. Further promoted people to analysis and research the images and video content. In the area of video content analysis, the researchers are particularly concerned about the video indexing and abstracting techniques. In the areas of image content analysis, researchers are particularly concerned about the automatic acquisition, analysis and monitoring of the scene content. In these two areas, the text showed its important role because it can provide directly high-level semantic information. Besides, the text can also be removed from the original image. When the background of the text area to be repaired. A lot of other lingual texts can be added. This has great significance in the exchange of images in different languages and the reuse of images.Text lines detection is actually a kind of pattern recognition. Unlike other typical pattern, the pattern of text lines is not fixed. At the same time the background of text lines tend to be more complex. All these have brought about difficulties to text detection. So the traditional method of detection of the image model can not be used to detect text lines directly. Based on research at home and abroad , two kinds of detection methods are proposed. The first algorithm is to use edge detection and local histogram to locate the text area. First, edge detection is used to locate the rough text areas. Second, local histogram is used to fixed accurate text lines. Last, structural information is used to remove the factitious text lines. The first algorithm can give quite good detection results if the background of the video is not very complex. The second algorithm is to use the internal structure of the text and video frame time redundant information to detect the text area. In the approach temporal information is introduced as the part of detection step to remove false positives. Finally as a post-processing step, density-based distribution method is employed to detect falsely detected features, that is, noises. The second algorithm also effect good.After the extraction of the text areas in the video sequences, as an application of the text detection, texture repairing technique is used to remove the text from the video.