Abrupt shot boundary detection with combined features and SVM
Youxian Zheng, Yuan Zhang · 2016
Abrupt shot boundary detection is a crucial pre-process of content-based multimedia processing tasks. On account of large object/camera movements and flash lights, global features are inadequate to detect abrupt shot boundaries. Hence, combined features of local and global features can be more robust. In this paper, an abrupt shot boundary detection framework with combined features and Support Vector Machine is proposed. The proposed framework jointly combines block HSV histograms and SURF to distinguish frames and adopts a proposed ensemble undersampling algorithm to process imbalanced data and uses SVM to automatically identify abrupt shot boundaries. The experiments show that the proposed algorithm achieves high accuracy.