Static Video Summarization using Sparse Autoencoders

Abstract

This paper presents a novel method for static video summarization. The video summarization methods generate an abstract representation of the input videos. The proposed method first extracts individual frames from the input video. Then, the redundant frames are eliminated based on motion vectors. The incorporation of redundancy elimination greatly reduces the complexity of the step in which the abstract representation is created. After the redundancy elimination, the high-level features are extracted from each frame. The feature extraction is done using Sparse Autoencoders (SAE). These high-level feature vectors from SAE are clustered using K-means algorithm. The frames closest to the centroid of each cluster is selected as the key frame of the input video. The method is tested on two benchmark datasets: VSUMM and OVP. The proposed approach attains better results compared to other state-of-the-art video summarization techniques. The efficiency of the proposed method is proved by comparing with other clustering-based approaches in video summarization.