Research On Text Detection From Video

Image texts play an important role for image content understanding since they contain rich semantic information. With the help of text, many tasks such as picture classification, index and retrieval become possible, and thus design all types of applications. Therefore, text detection becomes an urgent task in the recent years and it has important theoretical value and practical application prospects.Image content retrieval has always been a focal problem in multimedia and computer vision, and a number of algorithms have been proposed. Most of these methods have variations on specific image types, or too many prior hypothesis such as color, language, nosie and so on. Therefore, detection and segmentation of text from images robustly is still an unsolved problem, especially considering the existences of their degradations, noises, and complex backgrounds.In this paper, we explore research for text detection in video. We summarize the research progress for text detection and text recognition in the video and discuss typical methods in text detection. Then we propose a novel hybrid algorithm to fast detect video texts even under complex backgrounds. We first use an SVM classifier trained by our new StrOke unit Connection (SoIC) operator to identify seed stroke units. Then the stroke units are tracked and extended into their surroundings to form text lines, obeying seed stroke geometric constraints. Experimental results show that our approach is color and language independent, and robust to video illuminations. Further, we present a new framework to automatically detect video captions based on Optical Flow algorithm and our method is efficient for real-time video caption detection.