Comparison of video shot detection methods using higher order local descriptor

Video Shot Detection plays a vital role in the analysis of the contents in Video. The algorithms and methodologies learnt from Video Shot Detection has a wide range of applications starting from Video Browsing, Content-based Video Retrieval and Storage, Surveillance and many more. Video Shot Detection is the temporal segmentation of Video stream to determine the transitions of Video. Out of the two categories of transition viz., Hard Cut and Soft Transition, in this paper we propose methods to determine Hard Cut. Three Video Shot Detection Methods have been used in this work, they are Mutual Information, Weighted Variance, Likelihood Ratio and Edge Change Ratio. A comparative study has been conducted to find out the best Video Shot Detection method out of the three. The essential factor that drives the above three methods is extraction of appropriate features from the frames of video stream which would be used for the temporal segmentation of video to determine the transition. In this proposed paper we are using Higher Order Local Descriptors such as Local Binary Pattern(LBP), Local Derivative Pattern(LDP), Local Tetra Pattern(LTP) and Local Vector Pattern(LVP) and convert the original video into these feature videos. Frames from input video sequence are converted to texture domain and for each video sequence we generate four video corresponding to four Higher Order Local Descriptors. These video sequences are used to determine the `CUT' transition. Using QM Parameters, we found out the best feature among four Higher Order Local Descriptors for a given class of video.