The Simulation Of Several Visual Computation In Primary Visual Cortex And Their Application

Computer vision is of great importance to the research of intelligent computers. However, most of the researchers focus on the theories and algorithms proposed from the viewpoint of engineering. The neural mechanisms which the human visual system processes the visual information are ignored and the theories lack uniform algorithms and a complete computational system. Nevertheless, the human visual system possesses very excellent image processing abilities. So the study and simulation of human visual information processing mechanisms are possible to get models which are more in line with biological vision, promoting the development of the discipline of computer vision. In this paper, based on the neural mechanisms of retina, lateral geniculate nucleus, primary visual cortex and considering that the actual biological neurons use the time sequence of spike trains to encode information, some image processing methods are proposed. It mainly involves:1.We propose a hierarchical spiking neural network model to solve the edge detection problems. This model is based on the neuronal mechanisms at each level of the primary visual pathway. Concretely, according to the connection structure of ganglion eel 1, lateral geniculate nucleus cell, simple cell in the primary visual pathway and the Hubel-Wiesel connection model, we firstly sharpen the input image to realize edge enhancement by mimicking the spatial-opponent characteristic of the ganglion cell. Then, the edge enhanced image is delivered to the simple cells and the simple cells detect the edge in certain direction according to the Hubel-Wiesel connection model. (Based on the recognized processing methods in neurophysiology, we assume that the LGN cell just plays the role of relay and it has no special image processing functions. So our model does not include this layer.). Finally, the edge maps in different directions are combined to form the final edge image. Some model test results show that the proposed model is superior to the performance of some traditional edge detection algorithms such as Sobel, Roberts and it can also give comparable results as Canny.2.As one of the applications, we apply the model which imitates the neural mechanisms of double-opponent cells to the color correction of license plate. Then, we propose a spiking neural network model to detect the possible candidate areas of license plate. At the same time, the proposed edge detection algorithm is used to extract the edge corresponding to the candidate areas of license plate. The experimental results show that the bio-inspired algorithm of locating the license plate has excellent position accuracy rate.