Abstract: Dissolve is an important gradual transition in video，and it is the most difficult to detect in many different kinds of shot boundaries.The first difference of variance curve in successive frames is analyzed to extract candidate dissolve regions，and the character of successive monotonic change in dissolve region is used to exclude regions caused by motion and noise，which greatly improve the recall and precision.The algorithm is tested on comprehensive videos，and compared with Alattar method and Nams method，and the performance is proved to be more accurate and reliable.

摘要： 叠化是视频中的一种重要的渐变类型，也是各种镜头边界中最难检测的一种。首先利用视频连续帧序列的方差一阶差分曲线检测出候选搜索区域，然后根据叠化所具有的连续一致变化特征从候选搜索区域中排除由于运动及噪音造成的误检区域，从而大大提高了算法的查全率和查准率。为了验证本算法，实验中选取了大量视频，并且和算法Alattar及算法Nams进行比较，结果表明本算法具有较高的准确性和可靠性。