Real-time Caption Detection and Localization in TV Programs via Stroke Width Transform and Morphological Transform

Captions in TV programs are important clues for video content analysis, indexing, and retrieval. However, it is difficult to extract due to complex background, unknown caption color, etc. Nevertheless, caption detection and localization on its own is a hard task, let alone real time realization. In this paper, we propose a novel algorithm to extract caption from TV channel videos. Our proposed algorithm is composed of three steps: canny edge detection and binarization, simple stroke width transform, and histogram projection analysis. The experimental results show that the proposed method mentioned above is effective in real time TV caption detection and localization.