Video Summarization Based on ListNet Scoring Mechanism

With the increase of video data, we have a new challenge in the browsing, storage and retrieval of video, the presentation of video summarization technology is an effective way to solve such problems. In this paper, the convolutional neural network VGG19 model is used to extract the depth features of video frames on the TVSum50 dataset. The training set is trained by the sorting learning ListNet algorithm to obtain a model that can mark the frames of the test set. According to the scoring result, the top 25 frames in the test set are extracted as key frames, and the video summarization is synthesized. The video frames in digest which is came into being by automatically generated and manual work can use pHash algorithm as assimilation judgment, and get the F1-score of the model. Compared with the current video summarization method, this method is to mark the frame, not the frame set, so the calculation efficiency is reduced, but it can improve accuracy. Through the test on the TVSum50 dataset, the method of this paper is significantly better than other video summarization methods.