Video Summarization using Submodular Convex Optimization with Dynamic Support Vector Machine for Forest Fire Sequence Classification

摘要

This paper presents a new video summarization (VS) model, which summarizes the original and generally captured videos. The intention lies in the creation of precise summary which will convey the entire information. The summary includes the attractive and representative of the original video series. The earlier techniques are mainly based on simple considerations and optimizations. At the same time, they have utilized a hand-oriented objective which undergo sequential optimization by taking hard decisions. It restricts the usage in wide applicability. In this paper, a Submodular Convex Optimization (SCX) and dynamic support vector machine (DSVM) based VS model called Submodular Dynamical Video Summarization (SDVS) model is introduced. SCX is used for subset selection and DSVM is applied to classify the video summary. At the initial level, video sequence is given as input to the SDVS model. The transformation of input videos takes place to a set of frames. Next, extraction of key frames is carried out form the entire frame count for the generation of the video summary. For measuring the goodness of the SDVS model, a set of 8 videos are gathered from the Internet sources. The simulation outcome pointed out that the presented model achieved a maximum precision of 88.54, recall of 89.32 and accuracy of 88.91 respectively.