An algorithm for distinguish key semantic object from video

Distinguish semantic objects from the video data play an important role in content-based video analysis. Only a few semantic objects contain important information among these objects. We put forward the concept of key semantic object according to the human eye´s response to the features of video objects, and design the algorithm distinguishes the key semantic object among these objects. Firstly, we selected a combination of low-level features (e.g. luminance, shape, and rhythm) through which the semantic object can be characterized and distinguished; furthermore, the formulas for calculating the value of these features are given. Finally, the objects are filtered and sorted according to human visual characteristics of the video, the objects which can have a major impact on human visual system are distinguished. through analysis the different types of video, Experimental results show the proposed algorithm is suitable for the key semantic objects recognition of sports video and monitor video and achieves a high recognition accuracy rate exceeds 80 percent.