Video text localization based on Adaboost

Video text localization is an important step in video text recognition system for monitored control system, auto navigation system, content based image analysis system and so on. So it’s necessary to find a good method to extract text from video image accurately and quickly. In this paper a new method of video text localization based on Adaboost is proposed. Firstly, the edge detection is performed using Sobel operator based on the gradient feature to extract text region in the image. Through the analysis the feature of the video image region, five kinds of features are extracted to form five weak classifiers, then these features are classified to construct Adaboost strong classifier with CART (Classification And Regression Tree) and the candidates text regions are sent to the classifier to get correct result of the text region detection. The experimental results show that this method can not only achieve good effect on the text localization in the video images with the text of various fonts, sizes and colors, but also can realize rapidly and accurately to meet the video text localization requires.