Application of Frame Selection For Binary Video Classification

Abstract

This thesis presents frame selection based on genetic algorithm and Euclidean distance and its exploitation on binary video classification. The frame selection implementation provides a fast enhancement of classification results. Parallel frame selection methods are put up in comparison and a series of performance enhancement models are proposed to scrutinize the relationship with frame selection and binary video classification. Other approaches might also be valid to be exploited in the frame selection baseline in order to improve classification results. We could learn semantic meanings with the assistance of natural language processing, combine audio with pixel-level information to form a fusion model, or directly manage learning methods on video level.