一种基于小波的分层和多分辨的镜头边界检测方法

A hierarchical multiresolution video shot transition detection method based on the integer-to-integer wavelet

提出一种基于小波的分层和多分辨的镜头边界检测方法.在小波域上结合范数理论直接用小波系数构造相异性函数,并根据突变过渡检测中阈值难以确定的问题及渐变过渡易与运动等影响相混淆的特点,给出了方差比的自动阈值确定方法以及切比雪夫二次搜索策略,有效地克服了闪光和运动等的影响.用实际视频片段所做的仿真结果表明,该方法无论是突变镜头检测还是检测渐变镜头,检测率均高达98%和90%以上.

A detection method for the shot boundary detection is proposed based on the integer-to-integer wavelet, and hierarchical multiresolution. The dissimilarity function is presented by using wavelet coefficients directly combined with the norms theory, and according to the problem of threshold determination difficulty and the feature of confusing gradual transition with motion easily, the automatic threshold selection method and the scheme of twice search by the Tschebyscheff distance are presented. Integer-to-integer WT has advantages in speed over the first generation WT and adaptation to dealing with an amount of video data. This method can overcome flash and motion effects efficiently. The experimental results with real video clips demonstrate that a trade-off between recall and precision is up to more than 98% and 90% no matter what abrupt transitions or the gradual transitions.