Robust Video Text Detection with Morphological Filtering Enhanced MSER
Yun-Zhi Zhuge,Hu-Chuan Lu
DOI: https://doi.org/10.1007/s11390-015-1528-z
2015-01-01
Abstract:Video text detection is a challenging problem, since video image background is generally complex and its subtitles often have the problems of color bleeding, fuzzy boundaries and low contrast due to video lossy compression and low resolution. In this paper, we propose a robust framework to solve these problems. Firstly, we exploit gradient amplitude map (GAM) to enhance the edge of an input image, which can overcome the problems of color bleeding and fuzzy boundaries. Secondly, a two-direction morphological filtering is developed to filter background noise and enhance the contrast between background and text. Thirdly, maximally stable extremal region (MSER) is applied to detect text regions with two extreme colors, and we use the mean intensity of the regions as the graph cuts’ label set, and the Euclidean distance of three channels in HSI color space as the graph cuts smooth term, to get optimal segmentations. Finally, we group them into text lines using the geometric characteristics of the text, and then corner detection, multi-frame verification, and some heuristic rules are used to eliminate non-text regions. We test our scheme with some challenging videos, and the results prove that our text detection framework is more robust than previous methods.