A New Video Analysis Approach for Coherent Key-frame Extraction and Object Segmentation

We discuss a new video analysis approach for coherent key-frame extraction and object segmentation. As two basic units for content-based video analysis, key-frame extraction and object segmentation are usually implemented independently and separately based on different feature sets. Our previous work showed that by exploiting the inherent relationship between key-frames and objects, a set of salient key-frames can be extracted to support robust and efficient object segmentation. This work furthers the previous numerical studies by suggesting a new analytical approach to jointly formulate key-frame extraction and object segmentation via a statistical mixture model where the concept of frame/pixel saliency is introduced. A modified expectation maximization algorithm is developed for model estimation that leads to the most salient key-frames for object segmentation. Simulations on both synthetic and real videos show the effectiveness and efficiency of the proposed method