基于双重检测模型的视频镜头分割算法

A Video Shot Segmentation Algorithm Based on Dual-detection Model

提出一种基于自适应双重检测模型的视频镜头分割算法.该算法分为初检和复检两阶段,初检提出了基于人类视觉特征的非均匀分块的概念,并结合滑动窗口自适应二分查找算法进行镜头边界检测;复检过程利用尺度不变特征变换算法对初检得到的镜头边界进行匹配,排除初检中的误检.视频测试结果表明,该算法对多种不同类型的视频能同时提高切变和渐变镜头边界检测的效果.

Video shot segmentation is the fundamental work of content-based video retrieval.An adaptive shot segmentation scheme based on the dual-detection model was proposed.In the pre-detection round,the uneven blocking mechanism based on human visual features is presented and applied to adaptive binary search（ABS） method in moving window to detect shot boundaries.In the re-detection round the scale invariant feature transform（SIFT） algorithm will match the boundary frames and exclude false detections.The experiments indicate that this algorithm performs well on both abrupt and gradual shot boundary.