基于HEVC压缩域的镜头边界检测方法

镜头边界检测是智能视频检索的一个重要环节。现有的检测方法主要是在像素域进行处理,切变检测精度不高,计算复杂度过大。针对这些问题,文中利用解析HEVC码流得到的编码信息,提出了一种基于HEVC压缩域的镜头边界检测方法。首先统计每帧编码信息中各类预测模式的PU个数,并根据CU深度对运动矢量进行幅值滤波；然后采用PU预测模式、运动矢量和帧比特数对切变候选帧进行两级筛选,再对其进行自适应阈值的镜头切变检测；接着根据切变帧对视频序列进行分段,并在时域上对帧比特数进行平滑滤波；最后使用PU预测模式和经滤波平滑后的帧比特数对分段视频进行镜头渐变检测。实验结果表明,该方法具有良好的镜头边界检测效果,并具有较低的计算复杂度。

Shot Boundary Detection Method Based on HEVC Compressed Domain

Shot boundary detection is an important part of intelligent video retrieval.The existing detection methods are mainly processed in pixel domain,with low accuracy of cut and high computational complexity.To solve these problems,this paper used the encoding information obtained by parsing HEVC stream and proposesd a shot boundary detection method based on HEVC compressed domain.First,the number of PUs with different prediction modes is counted for each frame,and the motion vectors are filtered according to CU depth.Then,the two-stage candidate frame of cut is selected by using the PU prediction modes,the motion vectors and the number of frame bits.And then,the cut shot detection of adaptive threshold is performed.After that,the video is segmented according to the cut frame.In addition,the smooth filtering is carried out for the frame bits in the time domain.Finally,the PU prediction modes and the number of smoothed frame bits are used to detect the gradual shot detection.The experimental results show that the proposed method has a good effect on shot boundary detection with lower computational complexity.