This paper utilizes the physiological mechanism of non-classical receptive field of ganglion cell to design a hierarchical network model for image representation based on neurobiology. It is different from the contour detection, edge detection, and other practices using the classical receptive fields, simulating the non-classical receptive fields physiological mechanism which can be dynamically adjusted according to stimulation for image local segmentation and compression based on image neighborhood region similarity, thus realizes the inner image representation in neural representation level and convenients for extract the semantics of image further. We provide extensive experimental evaluation to demonstrate that, comparing with the traditional methods using single pixel, GC-array can represent image with a low cost. Most importantly, the GC-array model provides a basic infrastructure for image semantic extraction.