Performance analysis on video summarization using deep convolutional neural networks

Abstract

newline The quantity of video has emerged recurrently due to emergence in the fame and reduced video cost. It is termed as an imperative part of visual data. Hence, it is imperative to devise a model that can facilitate network browsing using huge data. Thus, the video summarization has gained huge importance to aids deal with large data. The video summarization models aimed to build a precise and best synopsis by choosing the best part of video content. Various methods are devised that relies on deep model. This research presents two contributions that are based on video summarization. The first contribution is to present a Mutual Probability-based K-Nearest Neighbour (MP-KNN) for video summarization. The videos with different frames are provided to keyframe removal wherein keyframe acquisition is performed with Discrete Cosine Transform (DCT) and Euclidean distance and then optimal keyframe is generated. The remaining frame is given as an input to Deep Convolutional Neural Network (DCNN). Similarity is computed with Bhattacharya distance. The optimum frameset is computed by matching the generated keyframes with residual keyframe. Thus, the input queries containing the face object are fed to object matching, and are executed with MP-KNN to produce pertinent frames with texture features.