Caption text location with combined features using SVM

News caption text contains useful information for video annotation, indexing and searching. This paper presents a new caption text location method. First, a small overlapped sliding window is scanned over the keyframe. Then texture and edge features are extracted as the input to SVM classifier to distinguish caption text from background. At last, vote mechanism and morphological filter are performed to precisely locate the caption text region. The new method is expected to outperform the existing strategies based on the following two improvements. One is to combine texture-based method and edge-based method to make the algorithm more robust to complex backgrounds and various font styles. The other is to address the multilingual capability over the whole processing. The proposed algorithm has been evaluated by four different TV channels and the experiments show its high performance.