A generic framework for Generation of Summarized Video Clips using Transfer Learning (SumVClip)

Video summarization aims to produce highlights of the original video showing informative key events. Now a days video content is increasing enormously therefor to store, browse and share videos is time consuming and a challenging task. A generic technique named SumVClip is proposed in this study. The generic video summary clips are produced by selecting key-frames based on classification through deep learning i.e. AlexNet. The model is trained for a dataset in different categories such as surveillance and lectures having two main classes positive and negative. To generate summarized video clips, threshold is set according to video length for generation of each video clip; hence frames are classified and added sequentially into the output video. For the evaluation of the proposed algorithm, various metrics i.e. accuracy, precision, recall and F1 scores is computed. The quantitative F1 score shows the encouraging results to generate summarized video clips.