Color Descriptors based Correlation Coefficients and Adaptive Threshold for detection of Wipe transitions

Shots are the basic elements which are used for video summarization and analysis. On the other hand special lightening effects, rapid object-camera motions and scene blending have posed significant challenges in content based video retrieval. A robust transition detection system may sometime be fooled easily owing to the consistency established between scenes and special effects during video editing. Wipes are gradual transitions which includes two simultaneous scenes spatially separated within the transition frames. We propose a simple and efficient approach to detect wipe transitions in videos subjected to uneven illuminations, poor backgrounds, object movements and camera motions. The scheme uses global feature obtained from sum of absolute histogram difference (SAHD) between two successive frames for each color component. The normalized value of mean of SAHD is then used to obtain an adaptive threshold. A two stage normalization correlation coefficient and the adaptive threshold with Expectation Maximization (EM) curve are then used to detect wipe transitions. The result over 27 videos shows that our approach was successful in detection of wipes in 26 videos.