改进的互信息量相似度曲线关键帧提取研究

摘要

关键帧提取是基于内容视频检索和比对等工作的关键步骤，它的准确性和效率是后续工作的基础。针对目前关键帧提取存在的阈值选择困难、算法复杂等问题提出改进的互信息量关键帧提取算法，构造基于非均匀分块加权HSV直方图的互信息量相似度曲线检测算法，取相邻两高曲率点的中间帧作为关键帧。实验结果表明，该算法无需阈值选择，能实现关键帧自动提取并且只需要一次扫描就能快速得到准确的关键帧。%Key frame extraction is a critical step in content-based video retrieval and comparison,its accuracy and efficiency is the basis of the successive work.To overcome the problems of difficulty in threshold selection and algorithm complexity existed in current key frame extraction methods,we propose an improved mutual information key frame extraction algorithm.It constructs a mutual information similarity curve detection algorithm which is based on non-uniform blocks weighted HSV histogram,and takes the middle frame between two adjacent high curvature points as the key frame.Experimental results show that the algorithm can achieve automatic key frame extraction without threshold selection.In addition,this method can also extract accurate key frames quickly by just single scanning.