PROJECT TITLE :

Multi-Level Discriminative Dictionary Learning With Application to Large Scale Image Classification - 2015

ABSTRACT:

The sparse coding technique has shown flexibility and capability in image representation and analysis. It is a robust tool in several visual applications. Some recent work has shown that incorporating the properties of task (like discrimination for classification task) into dictionary learning is effective for improving the accuracy. But, the traditional supervised dictionary learning methods suffer from high computation complexity when addressing massive variety of categories, creating them less satisfactory in massive scale applications. During this paper, we tend to propose a unique multi-level discriminative dictionary learning methodology and apply it to giant scale image classification. Our method takes advantage of hierarchical category correlation to encode multi-level discriminative information. Each internal node of the category hierarchy is related to a discriminative dictionary and a classification model. The dictionaries at totally different layers are learnt to capture the information of various scales. Moreover, each node at lower layers additionally inherits the dictionary of its parent, thus that the classes at lower layers will be described with multi-scale information. The learning of dictionaries and associated classification models is jointly conducted by minimizing an overall tree loss. The experimental results on challenging information sets demonstrate that our approach achieves excellent accuracy and competitive computation cost compared with other sparse coding ways for large scale image classification.