Eratosthenes sieve based key-frame extraction technique for event summarization in videos

TLDR: An Eratosthenes Sieve based key-frame extraction approach for video summarization (VS) which can work better for real-time applications and outperform the state-of-the-art models on F-measure.

Abstract: The rapid growth of video data demands both effective and efficient video summarization methods so that users are empowered to quickly browse and comprehend a large amount of video content. It is a herculean task to manage access to video content in real time where humongous amount of audiovisual recorded data is generated every second. In this paper we propose an Eratosthenes Sieve based key-frame extraction approach for video summarization (VS) which can work better for real-time applications. Here, Eratosthenes Sieve is used to generate sets of all Prime number frames and nonprime number frames up to total N frames of a video. k-means clustering procedure is employed on these sets to extract the key–frames quickly. Here, the challenge is to find the optimal set of clusters, achieved by employing Davies-Bouldin Index (DBI). DBI a cluster validation technique which allows users with free parameter based VS approach to choose the desired number of key-frames without incurring additional computational costs. Moreover, our proposed approach includes likes of both local and global perspective videos. The method strongly enhances clustering procedure performance trough engagement of Eratosthenes Sieve. Qualitative and quantitative evaluation and complexity computation are done in order to compare the performances of the proposed model and state-of-the-art models. Experimental results on two benchmark datasets with various types of videos exhibit that the proposed methods outperform the state-of-the-art models on F-measure.