基于组合相似性的视频检索

该文研究基于镜头的视频检索问题，提出了一种新的基于组合相似性的镜头相似性度量方法。首先把镜头看成由帧序列组成的一个组合，镜头的相似性通过帧组合的相似性来度量。其次通过用一个非线性映射，把帧组合所在的空间映射到一个高维空间，在这个空间中，假设帧组合服从正态分布，利用核方法，抽取出关键帧序列，并计算出两个正态分布之间的概率距离，这个距离表明了帧组合的相似程度，从而得到两个镜头之间的相似性。最后将这种方法应用于基于镜头的视频检索中，实验表明在相同条件下，基于该方法的检索效果明显优于传统的欧式距离和直方图交方法。

Ensemble Similarity-Blased Video Retrieval

In this paper, a novel method is proposed to determine the similarity between shots. Firstly, a shot is treated as an ensemble that consists of a sequence of video frames. Shot similarity can be measured by ensemble similarity. Secondly, the original space is mapped to a high dimension space by a nonlinear mapping. In this space, distribution of the ensemble can be assumed as a normal distribution. Finally, by kernel method, the probability distance is computed directly. This distance is equivalent to the ensemble similarity. So, the shot similarity is also obtained. Experimental results show that this method achieves superior performance than the traditional Euclidean distance and histogram intersection methods.