期刊名称:International Journal of Computer Science & Information Technology (IJCSIT)
印刷版ISSN:0975-4660
电子版ISSN:0975-3826
出版年度:2016
卷号:8
期号:1
页码:27
出版社:Academy & Industry Research Collaboration Center (AIRCC)
摘要:This paper introduces a novel approach for efficient video categorization. It relies on two maincomponents. The first one is a new relational clustering technique that identifies video key frames bylearning cluster dependent Gaussian kernels. The proposed algorithm, called clustering and Local ScaleLearning algorithm (LSL) learns the underlying cluster dependent dissimilarity measure while findingcompact clusters in the given dataset. The learned measure is a Gaussian dissimilarity function definedwith respect to each cluster. We minimize one objective function to optimize the optimal partition and thecluster dependent parameter. This optimization is done iteratively by dynamically updating the partitionand the local measure. The kernel learning task exploits the unlabeled data and reciprocally, thecategorization task takes advantages of the local learned kernel. The second component of the proposedvideo categorization system consists in discovering the video categories in an unsupervised manner usingthe proposed LSL. We illustrate the clustering performance of LSL on synthetic 2D datasets and on highdimensional real data. Also, we assess the proposed video categorization system using a real videocollection and LSL algorithm.