LOW-COMPLEXITY VIDEO RETARGETING BASED ON SMART CROPPING

A low-complexity video retargeting algorithm based on smart cropping is proposed. The YOLO-Fastest algorithm was used to detect video target, and the importance of prior box was determined according to area, accuracy, and type, to guide the selection of cropping window. The relative displacement of video frame was calculated according to the position of subject prior box, and the path of cropping window was constrained to protect temporal continuity of retargeted video. In this paper, a user study and two objective quality metrics (temporal continuity distortion measure and significant similarity measure) were used to evaluate retargeting performance, and time consumption was compared with existing algorithms. The experimental results show that the proposed algorithm can greatly reduce algorithm complexity and video retargeting time without obvious reduction of retargeting video quality.