| This dissertation reports on research, development, and evaluation of a color based video sequence identification and tracking algorithm. We describe an automatic video sequence identification and tracking algorithm that detects and extracts repeated video sequences from a continuous video stream. Because our technique is domain and video source independent, it is applicable to any video stream that is repetitive and changes slowly over time.; We digitize and segment a continuous video stream into video sequences using color histogram-based techniques. Our video sequence identification approach groups individual frames together based on their color features. In particular, we use the nine color moments, namely the mean, variance, and skew of each primary color component of the RBG color format. Then, we identify similar video sequences based on how many similar frames they contain. Finally, we compare the similar video sequences frame by frame to identify repeated video sequences.; We studied the efficiency and the effectiveness of our algorithms on 24 or more hours of video data from two different sources. We found that the technique accurately identifies repeated sequences, producing recall and precision values both over 90%. We also evaluated the applicability of our technique as a lossless compression algorithm. By removing repeated sequences from the video archive, we achieve a compression gain ratio of 30% on each source. |
