A novel keyframe extraction method for video classification using deep neural networks

Abstract Combining convolutional neural networks (CNNs) and recurrent (RNNs) produces a powerful architecture for video classification problems as spatial–temporal information can be processed simultaneously effectively. Using transfer learning, this paper presents comparative study to investigate how temporal utilized improve the performance of when CNNs RNNs are combined in various architectures. To enhance identified effective combination CNN RNN, novel action template-based keyframe extraction method is proposed by identifying informative region each frame selecting keyframes based on similarity between those regions. Extensive experiments KTH UCF-101 datasets with ConvLSTM-based classifiers have been conducted. Experimental results evaluated using one-way analysis variance, which reveals effectiveness sense that it significantly accuracy.