SIMULATING HIERARCHICAL STRUCTURE OF HUMAN VISUAL CORTEX FOR IMAGE CLASSIFICATION

Abstract

In this thesis, we study several hierarchical models for image classification that are biologically inspired and simulate some known characteristics of visual cortex. We base our investigation on the HMAX model, and extend this model in several aspects such as adding clustering of features, evaluating different pooling methods and coding occurrences and co-occurrences of features, with the goal of improving the image classification accuracy on benchmark datasets. We further propose a new high-level biologically inspired color model, CQ-HMAX, which can achieve better performances than the state-of-the-art using the bottom-up approaches when combined with other low-level biologically inspired color models and HMean and use these models for detection of mitosis in histopatholgy images and classification of benthic marine organisms. We also propose an HMAX like structure for simulating auditory cortex and create sonar images and combine them with visual images for underwater image classification in poor visibility conditions.