A Stochastic Framework for Keyframe Extraction

EasyChair Preprint 2753

Abstract

The sudden growth in the Closed Circuit TeleVision (CCTV) installations has paved the way for intensive video analytics. Video summarization, being a method of representing keyframes of a voluminous video, plays a major role in the video processing. Several researchers have focused on the key frame extraction since late 90’s. However, several challenges still exist in keyframe extraction. The main challenge in keyframe extraction is to identify the representative frames based on the contents. Most of the existing methods adapt deterministic approaches, which involves more computational complexity and result in poor accuracy. This work aims to improve the accuracy rate by introducing a stochastic framework that uses the techniques such as binning, Markov chain, Transition Probability Matrix (TPM), and Permutation computation. The experimental results demonstrate that the proposed framework outperforms the existing methods VSUMM and VSUKFE in terms of both
computational efficiency and accuracy.

Keyphrases: Markov chain, Permutation computation, keyframe extraction, stochastic framework, transition probability matrix, video summarization

@booklet{EasyChair:2753,
  author    = {Thangaswamy Judi Vennila and Vanniappan Balamurugan},
  title     = {A Stochastic Framework for Keyframe Extraction},
  howpublished = {EasyChair Preprint 2753},
  year      = {EasyChair, 2020}}