摘要： 根据HSV色彩直方图进行镜头边界检测是一种常用、有效的方法，该文提出了一种基于色彩和形状融合的方法计算帧间不连续值、形状特征选取梯度方向角，在计算出帧间不连续值以后，采用Kohonen自组织网络对不连续值进行聚类得到镜头边界。实验结果表明该方法能弥补采用单一的色彩或形状信息所造成的漏检或错检，并且Kohonen网络能取得与滑动窗高斯模型方法相差无几的性能表现，却克服了检测性能对于参数的敏感性。

关键词: 镜头边界检测, 梯度方向角, Kohonen网络, 基于内容的视频检索

Abstract: HSV histogram is a commonly used, effective method for shot boundary detection. This paper deals with computing discontinuity values based on shape and color content in image frames. Then Kohonen self-organized network is used to differentiate boundaries from frames. Experimental results show that this method can avoid wrong and omited detection as a result of adopting single characteristic, and Kohonen net can perform as well as slide window Gaussian’s model method, while diminish the influence of the parameters on the detection performance.

Key words: shot boundary detection, gradient angle, Kohonen network, content-based video retrieval

中图分类号: 

• TP18