Efficient Shot Boundary Detection with Multiple Visual Representations

Table 3

Average precision and recall values of input videos.

Video input	Total no. of cuts	No. of cuts detected	Proposed method
	TP + FN	TP + FP	Recall	Precision	F1 score
V1	101	99	0.97	0.99	0.979
V2	81	82	0.975	0.98	0.977
V3	106	105	0.98	0.99	0.985
V4	91	93	0.978	0.98	0.979
V5	96	98	0.98	0.98	0.98
V6	76	75	0.94	0.933	0.936
V7	119	118	0.96	0.98	0.97
V8	216	216	0.972	0.972	0.972
Total/Average	886	886	0.969	0.976	0.972

Table 3 shows the recall, precision, and F1 score values of selected input videos. The total number of actual cuts, number of cuts detected by the proposed method, and average values of recall, precision, and F1 score achieved by the proposed method are shown in bold. These score shows the better performance of our SBD method. This high performance is achieved by the use of multiple invariant features and the SVM classifier. As we have used important features such as color feature, edge feature, and motion feature, we could overcome most of the variance such as illumination changes, rotation variance, scaling, motion effects due to camera motion, and object motion.