A general image representation scheme and its improvement for image analysis

In this paper, a bio-vision neural network was presented could provide a novel approach for image representation and a processing method for image analysis. This model using ganglion cells (GC) in retina and its non-classical receptive field (nCRF) can dynamically self-adjusts based on different input images. We provide extensive experimental evaluation to demonstrate that, this GC array can represent image with a low cost and could substantially improve image processing efficiency. Most importantly, the image representation based on GC-array model provides new approach for image semantic extraction later.