Opposition-based optimized max pooled 3D convolutional features for action video retrieval

Abstract

Key frame selection serves as a c bridge between raw video data and meaningful retrieval results. Effective key frame selection enhances the performance of content-based video retrieval systems by reducing computational complexity, improving search accuracy, and enabling faster browsing through large video databases. Additionally, fixed keyframe sampling techniques do not address information optimization, which might lead to information redundancy or loss. For effective video retrieval, a keyframe selection method based on opposition-based learning has been developed. The outcomes show that the method performs better than numerous benchmark sampling strategies.