A Robust Text Segmentation Algorithm in Images and Video Frames

摘要

In this paper, an algorithm for text segmentation in images and video frames is proposed. First, we rescale text blocks detected to a suitable size and exploit a Gaussian filter to remove noises. Then a procedure of flood filling from border pixels follows to remove most of complex background. After that, those remaining pixels are clustered into several layers through a modified k-means algorithm and several heuristic rules are adopted to help locate the most probable text layer. Finally, some post-processing techniques are designed to further improve the segmentation results. The experimental results show that the proposed algorithm can automatically and effectively segment text from images and video frames with complex background, and is robust to text color, font size, as well as different styles of characters from different languages.