基于BIC的新闻视频近似重复帧检测方法

Near-duplicate keyframes detection in broadcast domain based on Bayesian information criterion

近似重复帧检测是新闻视频检索和追踪的重要组成部分。简要介绍了近似重复帧的定义,给出了所使用的角点检测方法及优点,提出了使用BIC对两个帧的特征值序列进行判决来确定是否近似重复的方案。实验结果表明,该方法在不需要设定阈值和机器学习的同时,召回率和准确率都有很好的表现。

Near-duplicate keyframes detection was an important part in content-based broadcast video search. After the definition of the acoustic change detection in audio segmentation was introduced, a new method based on Bayesian Information Criterion （BIC） algorithm was proposed for detecting near-duplicate keyframes, which result from the feature of salient parts extracted from a frame. Experimental results show that this method without threshold and learning has good performance in recall and precision.