Video Retrieval Based on Motion Vector Key Frame Extraction and Spatial Pyramid Matching

Video retrieval technique aims at retrieving relevant video from a large video dataset. Generally, it consists of heterogeneous videos and retrieval of relevant videos with high accuracy is a tedious work. In this paper, we proposed a content-based video retrieval system that correspondingly retrieves the most relevant videos from the database based on the visual content. In this context, motion vector based key frame extraction is computed as video summarization technique to recapitulate the video content and spatial pyramid matching scheme is used for the key frame matching. Computational mechanism for key frame selection is based on the concept of shot selection using outliers detection followed by sub-shot detection for each shot using motion vector. Finally, in each sub-shot undulant and stable sub section. Spatial pyramid matching partitions the key frames into increasingly fine sub-regions and computes features from each sub-region. The proposed method has been implemented and tested on real datasets. The performance has been compared with other standard methods. Experimental results provide an insight of satisfactory results in video retrieval, in terms of both subjective visual perception and objective evaluation metrics like fidelity, video sapling error, precision and recall.