一种鲁棒的视频指纹提取和匹配方法

摘要

Since existing some defects of the current video fingerprint approaches including weak robustness, ineffective matching scheme and limited adaptability, it presents a robust method for video fingerprint extracting and matching.The video fingerprint vector,which is combined the improved Harris descriptor with the improved inter-frame difference signature, is introduced.The hybrid video fingerprint matching scheme is designed for real-time working.The experimental results verify that comparing with the current method,the proposed video fingerprint is more robust to resist various geometric and signal based modification.%在视频广告监测的解决方案中,视频指纹日益受到重视.针对目前视频指纹鲁棒性不强、匹配方法效率不高或适用度不广的缺点,提出一种快速鲁棒的视频指纹算法:在视频指纹提取阶段,结合改进的Harris检测和改进的帧间差异法分别提取图像特征、运动特征以产生鲁棒的视频指纹；在指纹匹配阶段,设计了一种分层方案,该方案结合了典型匹配算法适用度广、位置敏感哈希(LSH)高效性的优点.实验表明,在亮度变换和尺寸变化以及常见的噪声攻击中,提取的视频指纹有较强的鲁棒性,提取、匹配的效率高.该方法满足视频监测准确性及实时性的要求.