一种基于支持向量的镜头聚类算法

Support vector-based shot clustering algorithm

对支持向量聚类中核区域的形成原理进行了深入分析,阐明了核区域在支持向量聚类进行重叠数据处理时的独特作用。针对视频数据内容存在大量数据重叠分布的特点,提出了一种基于支持向量的镜头聚类算法。利用颜色和时间作为特征向量,计算特征空间的聚类核区域,进而产生镜头聚类,克服了传统镜头聚类算法计算量大、仅以时间阈值判断镜头相似度等缺陷。

The formation theory of the core of support vector clustering was analyzed deeply, which disclosed the particular role of the core in disposing the overlapping data by support vector clustering. And then, a support vector based shot clustering algorithm was proposed according to the features that there are many overlapping data in the video sequences. In the algorithm, color and time were used as feature vectors to calculate the clustering core of feature space, further more, to form shot clustering. The algorithm addresses the problems of traditional shot clustering algorithms such as large amount of computations and using time as the unique threshold to estimate the similarities of shots.