Web Video Data Clustering and Recognition Using Histograms of Phoneme Symbols

Abstract

The clustering and recognition of Web video content play an important role in multimedia information retrieval. This paper proposes a method for both clustering and recognizing Web video content using a histogram of phoneme symbols (HoPS). HoPS contains information about speech and sound intervals. In this study, three experiments were conducted. The first experiment allocated HoPS feature of video intervals in a 3D space using PCA and quantification method IV (Q-IV). The second experiment applied the k-nearest neighbor (k-NN) method to analyze the difficulties in clustering. The third experiment recognized unknown video intervals by using the distance between HoPS of the query and a category average. The accuracy of the recognition results were 44.3% and 36.9% using the Mahalanobis distance and the correlation distance for the category average of training data, respectively.