A Neurocomputing Model For Image Cognition Based On Ganglion Cell Receptive Field’s Color Opponency Mechanism

After thousands of years of biological evolution, creatures struggle with the law of "survival of the fittest" and make best use of their advantages and also bypass the disadvantages, trying to grow up healthily. Human beings, as one of higher animals, have many fantastic and advanced abilities. For example, we humans can easily and quickly understand the complex scenes which are in front of us.Computer vision and artificial intelligence have always been a very important topic in the field of computer science. Researchers are dedicating themselves to solve problems which are related to image processing, such as how to make computers process, extract, segment, understand and recognize image intelligently and effectively, just like humankind. However, an image itself contains a great volume of data, which are also very complex. It also has strict requirements on the computing speed, processing efficiency and storage capacity. What is worse, currently most popular image processing methods are based on the individual pixel in the image, ignoring the basic property of the image, such as the strong relationship between each pixel. And a great majority of these algorithms are built purely on mathematic or statistic methods, neglecting the brain mechanisms and visual theory of physiology.Fortunately, with the development of biology and brain science, people have a better understanding on the complex nervous system. Nowadays, computer scientists also note that they can introduce and integrate some important mechanism of human visual perception system and model into the computer vision research, improving the efficiency and accuracy of image processing.This paper designs an image analysis and representation algorithm, which is based on biological visual theory and mechanism. We establish a neural network computing model for the retinal ganglion cells and their biological classical receptive field. This model involves the color opponency theory and simulates the eye micro fixation. The receptive field will scan and analyze the area it covers to adjust itself according to the characteristics of the image, the connection between pixels nearby and the feedback of the stimulus to the image. For the area where the characteristics are similar, the receptive field will expand to be relative large to cover, scan, analyze and represent. While for the area where there are a lot of differences, the receptive field will shrink to be relative small to cover, scan, process and represent. In this way, the essential contents of the image would be grasped while the reduntant ones would be ignored and all the receptive fields will stay in the most suitable status to make best of themselves, achieving excellent results.Experimental results show that this neuro-computing image cognitive model for ganglion cell’s receptive fields with the color opponency mechanism can represent the image accurately and effectively, and the representative results are loyal to the original images. It can reserve the crucial structural information of the image and suppress the trivial information at the same time. On the basis of these new representations, some up-coming image processing operations, such as image segmentation, could be greatly improved.At last we analyze the current model to find some points which can be improved and also look into the future on the topics which we can dig into and go further, hoping that we can achieve new highs in the field of biological visual simulation and computer modeling.