Abstract: There are two difficulties to recognize the text images which are extracted from videos: 1) how to segment and recognize the merged characters; 2) how to segment and recognize the characters with complex backgrounds. To overcome the difficulties, a novel integrated segmentation and recognition method is proposed. The method first binarizes the text image and estimates the height of the text line. Second, the connected components in the binary text image, which are wider than a threshold, are segmented based on image analysis or character recognition. Third, the connected components are selected and combined to generate the character patterns based on character recognition. Last, the best character sequence is selected based on a statistical language model. Experimental results demonstrate the effectiveness of the proposed method.