A robust and lightweight feature system for video fingerprinting

In this paper, a new content-based feature identification method for video sequences is presented. It is robust to a number of image transformations and relatively lightweight compare to most state of the art methods. A scale and rotation invariant descriptor for a set of interest points in detected key frames is proposed based on modified minimal spanning tree algorithm. In addition, a predicative coding scheme is used to achieve minimal size of the descriptor for transmission. Furthermore, the pairwise distance between the frequency responses of the curvature vector from the descriptors is calculated and compared to efficiently match query with a large database. Experimental results demonstrate the effectiveness of our approach.