A Near-Duplicate Video Cleaning Method Based on AFENet Adaptive Clustering

In recent years, with the growth of video data, problems such as video quality degradation and copyright protection difficulties have arisen due to the redundancy of a large number of near-duplicate video data. Aiming at the problems of insufficient video feature extraction and low accuracy of clustering algorithms in existing near-duplicate video cleaning algorithms. This paper proposes a near-duplicate video cleaning method based on AFENet adaptive clustering. The method in this paper mainly includes two parts: Firstly, in the feature extraction stage, this paper adopts the spatiotemporal features of video and enhances the features of salient objects in each frame of image, enhancing the feature expression ability of video; Secondly, in the video cleaning stage, an adaptive feature distance threshold SFD-Means clustering algorithm is proposed, and a clustering optimization function is designed to overcome the defect of the original FD-Means algorithm that is too sensitive to threshold selection. Experimental results show that our proposed method has high recognition accuracy and cleaning performance, effectively improving the quality of video data.