Video Key Frame Selection Based On JSD

With the rapid development of digital technology and network technology, digital video in our life is becoming more and more popular, people can easily use their digital equipments shooting and editing digital video. And there also emerged a large number of professional online video websites, large video databases are also becoming more common. Due to the current rapid growth of digital video, there is a growing urgent need for video summarization techniques to help people quickly learn and understand so many digital videos. Video summarization techniques produce summaries by analyzing the underlying content of a source video stream, condensing this content into abbreviated descriptive forms that represent surrogates of the original content embedded within the video.As a kind of video summarization technique, key frame extraction method is particularly important. The key frame selection is aimed to extract a set of typical frames that represent the visual content of a video sequence; the graphical presentation of key frames helps understand the video content. Key frame extraction is a basis of video summarization, provide a summary of the key frame information to different types of summarization methods; Meanwhile, the key frame information forms the data basis for video retrieval, indexing technology. By the using of key frame, we can greatly streamline the original video content with small storage space requirements, and it's very convenient to be used on a large video database.In this paper, key frame extraction techniques are summarized, several common methods are analyzed, the core of key frame extraction method: the frame distance and the logic of selection are described in detail, the evaluation criteria of key frame extraction merits of both objective and subjective are introduced. Based on this basis, we propose a generic method for extracting key frames in which the Jensen-Shannon divergence (JSD) is employed to measure the difference between successive video frames to segment a video clip into shots, and to choose key frames in each shot. The novel method for key frame extraction is computationally inexpensive and yet effective, as shown by experimental results.