Image Feature Extraction And Object Recognition Based On Visual Neural Mechanisms

Processing of visual information has long been one of the challenges in artificial in-telligence and computer vision. Despite such complexity, human brains manage this task accurately and effortlessly. Building a system that emulates the neural mechanism is al-ways an attractive but elusive goal. In the past decades, new methodologies have brought explosive growth to the study of neural mechanism underlying visual processing. The use of visual neuroscience in computer vision has been increasing accordingly.In this paper, we build computational models for simple cells and complex cells in the primary visual cortex. Furhter, we investigate into higher stages in the ventral visual pathway, which contributes to object recognition in the brain. Different stages extract visual features in different levels such as edges, orientations, contours and shapes. We build a multi-layer neural network model to emulate the visual pathway.We demonstrate experimentally that local image features generated by our model ex-hibit robust discrimination and even better generalization ability compared with traditional local features. We also evaluate our model in object recognition and localization experi-ments. This model provides a framework to emulate the mechanism of the ventral visual pathway. As the study of visual neuroscience advances, this model will continue to evolve.