Video fragment format classification using optimized discriminative subspace clustering

Abstract

All video streams consist of highly compressed coded data. A video stream must be decoded to identify a video. It is impossible to decode and identify a video fragment without knowing the correct video format. Therefore, the first issue that must be addressed is classification of video formats. Although several methods exist for classifying file formats, a technology that specifically classifies the formats of video fragments has not been developed. In this paper, we present a novel approach to classify the formats of small fragments of video streams. Our classification procedure involves construction of high-dimensional feature vectors by combining synchronization patterns extracted from training fragments. The feature vectors are classified using optimized discriminative subspace clustering (ODiSC). The experimental results show a minimum classification error rate of 4.2%, and the precision of identification of the formats was greater than 91% for the four video formats whose fragment size was 256KB.