A study on Genetic Algorithm based video abstraction system

This paper proposes to investigate on the efficacy of Genetic Algorithm (GA) based video abstraction system to deliver a meaningful summary (still image abstract) with minimal preprocessing on the given video. The GA employs novel crossover and mutation operators to search for a meaningful summary in a search space of all video summaries. This preliminary investigation employs Euclidean and City-block distance measures, based on simple color histogram, color histogram by Gong and color correlogram, among the frames as fitness functions and assume that the number of frames in the still image abstract is known apriori. The performance of GA based video abstraction system has been tested on 6 documentary videos from the open video project. The simulation results, though not very promising, strongly hints the potential of GA for automatic video abstraction and motivates further exploration.