Key frame extraction algorithm for video summarization based on key frame extraction using sliding window

The explosion of video content online makes finding specific information a challenge. Existing key frame extraction methods struggle to keep up with the variety of video formats and editing styles. This paper proposes CGSW-KF (Combined Gist Sliding Window Key frame), a novel key frame extraction algorithm that tackles this challenge. CGSW-KF leverages the strengths of SURF (Speeded up Robust Features) and GIST (Global Image Structure features) within a sliding window framework to accurately identify important frames. We use Dynamic Negative Sampling (DNS) to refine key frame selection, leading to a more focused and informative set of key frames. We evaluate CGSW-KF on a public dataset, demonstrating that it achieves competitive performance with deep learning models while offering better efficiency and interpretability. Our findings demonstrate the efficacy of CGSW-KF in improving video search, summarization, and indexing, hence enabling smooth navigation in the growing multimedia environment. We find an increase of 2.49% points over the state-of-the-art (SOTA).