A Framework for Background Detection in Video

This paper presents a framework for background detection in video. Key frames are extracted to capture background change in video and to reduce the magnitude of the data. Then we analyze the content of the key frames to determine whether there is interesting background in them. Key frames are extracted with a time-constrained clustering algorithm. Background detection in a key frame is done with color and texture cues. The illumination varies much in natural scenes. To deal with the varying illumination, color is modeled with three sub-models: strong light, normal light and weak light. The connectivity of background pixels is used to reduce the computing cost of texture. Experimental results show that background can be detected by using the framework simply and efficiently.