Multi-oriented moving text detection

Detection of moving text of different orientations in video is challenging because of low resolution and complex background of video. In this paper, we propose a method based on motion vectors to identify the moving blocks which have linear and constant velocity. For each block, we compute moments and use k-means clustering algorithm to extract text candidate. We introduce a new criterion based on gradient direction of pixels in text candidates to remove false text candidates which we outputs potential text candidates. Then the method performs region growing to group the potential text candidates which outputs text lines. The method is tested on both static and moving text video to evaluate the performance in terms of recall, precision, F-measure, misdetection rate and time. The results are compared with the well-known existing methods to show effectiveness of the proposed method.