在静态视频摘要生成过程中,提取代表帧是关键。通过分析视频帧底层的颜色特征,根据其内容自动设置连通阈值,提取颜色聚合向量,并以此进行基于等价关系的自适应聚类。在确定整体划分后,根据时序特征进行局部修正。整个过程无需设定任何阈值和人为干预,对变化程度不同的各种视频进行实验,结果比较满意。
It is very essential to extract representative frames in the process of generating video summary. A method is proposed which analyses the color features of the video frames, sets the connectivity threshold values automatically according to the contents, extracts the color coherence vectors (CCV) and then performs adaptive clustering based on equivalent relation. After global partition, local partition was revised with time sequential features. The whole process does not need to set any threshold values. The experiments with diverse videos yields effective results.