MULTIPLE CLASS MULTIPLE-INSTANCE LEARNING AND ITS APPLICATION TO IMAGE CATEGORIZATION

• DOI: 10.1142/s021946780700274x
• Publication Date: 2007-07-20T01:01:52Z
• Authors: XINYU XU; BAOXIN LI
• Keywords: Feature vector; Binary classification; Contextual image classification

We propose a Multiple Class Multiple-Instance (MCMI) learning approach and demonstrate its application to the problem of image categorization. Our method extends binary for Instead constructing set classifiers (each trained separate one category from rest) then making final decision based on winner all classifiers, our directly allows computation multi-class classifier by first projecting each training onto feature space simultaneously minimizing objective function in Support Vector Machine framework. The is constructed instance prototypes obtained which treats an as instances with labels being associated images rather than instances. experiment results two challenging data sets that achieved better classification accuracy less sensitive sample size compared traditional one-versus-the-rest MI methods.