ABSTRACT
This chapter discusses a new framework is proposed for video concept classification with the help of discriminative learning and multiple correspondence analysis (MCA). It outlines a new dual-model discriminative learning framework for video semantic classification is proposed to address the challenges such as semantic gap, imbalanced data, and high-dimensional feature space in automatic multimedia semantic analysis. Several new features are being developed to try to capture the semantic meaning of a video. The correlation and reliability information generated from MCA are utilized to select two discriminative sets of features, one set for the positive class and the other one for the negative class. MCA constructs an indicator matrix with instances as rows and categories of valuables as columns. MCA is designed for nominal data but has been proved to be able to effectively capture the correlation between a feature and a class.
