Video Classification And Retrieval Based On Convolutional Neural Networks

With the great success achieved by deep learning in image recognition research area,deep learning becomes more and more important in other research areas,including video classification,natural language processing,audio classification.One of the most important models used in deep learning is Convolutional Neural Network.Compared to the traditional neural network,deep neural network contains new layers and use new optimizer to train the model.After it's widely used in image recognition,the researchers begin to apply the Convolutional Neural Network to the video classification task.This paper also use the hash algorithms came up with by researcher in image retrieval research area.One part of this paper is about video classification.The traditional video classification algorithm is based on the features extracted from key frames.But now,the deep learning is becoming a more powerful tool to solve this problem.The mainstrea algorithm combines the Convolutional Neural Network and Recurrent Neural Network.The other part of this paper is about video retrieval.This paper use the algorithm.which is used in image retrieval,and try to prove it is workable in video data also.After analyzing the research work in related areas,we choose to use the video classification model based on VGG-NET which is one of the most successful Convolutional Neural Network models,and add a full-connected layer with Sigmoid function as activation function which is used to get binary hash code for each video.After we have got the binary hash code for each video,we come up with a new algorithm based on binary code and Trie to search for the similar videos.Compared to basic algorithms,it can improve the time used in similar videos searching effectively.