Contour Detection Algorithm Based On Non-classical Receptive Field Properties

Edges and contours are of great information of an image, and this information is very important in computer vision, image processing and machine intelligence. Also it is a big problem in these relevant fields. For one reason, these are so many textures and so complex background imbedded in.Human vision system can easily handle these problems, it can extract object from a very complex scene, and edges and contours are firstly detected in a very early stage of vision information processing. For this mission, the receptive fields play a very important role. By research the receptive fields'mechanisms and apply them to the image processing, the edge and contour detect process will be remarkably improved also the result will be much suitable for human vision characteristic.In this thesis, the traditional edge detectors were firstly discussed, some representative ones'theories were illustrated. Based on the illustration, the different between edge and contour were drawn. Then a new contour detector algorithm based on the orientation selective inhibition property of non-classical receptive field was proposed by use the mechanism of non-classical receptive field orientation selectivity inhibition in contour detector. Another contour detector base on the clustering mechanism of the receptive field in primary visual cortex neurons was proposed by importing the cluster mechanism of homogeneity. Experiments were done by applying both methods and the traditional detector to some natural images. The result shows great improve with these two detectors. To further verify the effect, a quantitative computation was introduced. And the data proved the improvement of the detector.