Rolling and non-rolling subtitle detection with temporal and spatial analysis for news video

Abstract

In this paper, We have developed a rolling and non-rolling subtitle detection algorithm with temporal and spatial analysis for news video. This paper mainly includes rolling subtitle region detection, character segmentation for rolling subtitle, and non-rolling subtitle detection and location. While, most previous video text detection methods aimed at non-rolling subtitles, this paper proposed a detailed detection algorithm for both rolling and non-rolling subtitles in currently popular news video. The proposed algorithm makes good use of typical features of rolling subtitles, and is efficient for rolling character segmentation. For non-rolling subtitles, it utilizes edge features and connected component analysis of subtitle regions. In the experiments, we have tested the proposed algorithm with a series of news video and achieved good detection and segmentation results.