Research on Video Information extraction algorithm based on Deep Learning
Download as PDF
The emergence of a large number of video data shows a greater demand for video abstract, and the existing feature-based and shot-based video abstract extraction methods are difficult to meet the actual needs in terms of computation, accuracy and reliability. using multi-feature layering of video to segment shots, using the strategy of first coarsening and then fine, using simple feature segmentation and then clustering complex features. Accurate video clips and key frames are obtained, and then global features are extracted from each key frame, and the similarity is compared to generate the final video abstract. The video summary is generated adaptively without considering the weight of multiple features. The experimental results on public video data sets such as VSUMM show that the multi-feature layering method effectively improves the performance of video abstract extraction. The accuracy and recall rate are better than the traditional methods, and the computational complexity is obviously reduced.
Video abstract; hierarchical clustering; multi-feature similarity; video segmentation; key frame