关键词: 视频存储, 视频检索, 深度学习, 卷积神经网络, 重新训练技术

Abstract: In order to improve the efficiency and accuracy of video retrieval, a schema of key frame abstraction and retrieval of videos based on deep learning is proposed. Firstly, an adaptive key frame selection algorithm is designed, and the distances of wavelet transforms are used to detect the frames belong to the same shot; then, abstract information of each shot is abstracted, and the frames containing the most significant features are set as the key frame of the corresponding shot; lastly, the existing convolutional neural network framework is used to abstract the features of key frames, and unsupervised, semi supervised and supervised retraining models are designed to improve the effect of the feature abstraction of the convolutional neural network and the accuracy of the video retrieval. Experimental results based on the public video datasets show that the proposed schema realizes a good precision for video representation, and it realizes an accuracy and high efficiency video retrieval too.

Key words:

 Video storage, video retrieval, deep learning, convolutional neural network, retraining technique