An Efficient Video Summarization Using Sparse Auto-encoders and K-means Clustering on Video Similarity Graph

This paper presents a novel approach for video summarization integrating sparse auto-encoders and K-means clustering based on the video similarity graph (VSG). The suggested approach begins by separating each frame from the input video. Then, each frame is augmented by incorporating some minor changes. Sparse auto-encoders (SAE) are used to create a high-level feature map of every frame following aug-mentation. Afterwards, K-means clustering based on the video similarity graph of feature vectors is used to create the clusters so that each cluster has frames with similar features. Finally, the most informative video summary is created by extracting key-frames from each cluster. We tested the proposed method on a single standard benchmark SumMe dataset, and the findings undeniably show how much better the given approach is than several state-of-the-art techniques.