A new video text extraction approach

Video text provides precise and meaningful information about video content. Video text extraction is a crucial step to retrieve video text characters. Most of papers perform video text extraction in a whole text row. Compared with whole text row extraction, single character extraction can achieve higher accuracy because the background of single character is relatively simple. However, character segmentation is difficult because of the low resolution of the characters in video. Therefore, we propose a character segmentation method, which can accurately locate the character boundary in the text row. In the text extraction, we perform k-means clustering in single character and define a method to quickly select one of the images as the final binary image. Experimental results on TV news video show the encouraging performance of the proposed algorithm.