A new method to extract Uighur text region in images

In view of Uighur text characteristics being stable and its background being changeable and complex in Video Frames, the thought of target attention is used to solve the problem of detecting Uighur text region in image. First, a method of color reduction is adopted to remove color gradient of text region. Based on this, suspected Uighur character regions are selected out of images with complex background by features of Uighur character region(focused goal 1), in the primaries, which are many small connected regions and are served as suspected samples. Second, a group of characters reflecting the relations (focused goal 2) between adjacent regions are extracted and one of characters is leading. Then every character´ information is combined into the leading character and a progressive cluster algorithm based on leading character is proposed. Finally, Uighur text regions in images are obtained with the aid of layer features (focused goal 3) of Uighur text region. Experiments show that the region of Uighur´s scene text and caption text on complex background can be segmented successfully. The accuracy rate is 98% and false alarm rate is lower than 14%.