Research On Contour And Boundary Detection Algorithm Based On Human Visual System

Contour and boundary detection is a fundamental research in the field of computer vision, and has been widely studied in recent decades, because of the complexity of natural images, contour and boundary is difficult to be detected, which will directly affect advanced visual tasks. In this thesis, we focus on the research of contour and boundary detection in natural images, introducing the human visual information processing mechanism we propose two contour detection algorithm: one is contour and boundary detection based on the combination of receptive filed and orientation modulation, the other is contour and boundary detection based on visual color theory and homogeneous suppression.First of all, we had a study on contour and boundary detection based on isotropic suppression of non-classical receptive filed. The 2D-Gabor function was used to simulate the classical receptive field of the simple cell in primary visual cortex(V1), the DOG function was used to simulate the non-classical receptive field, compared with Canny operator, the algorithm can retain contour and suppress texture edge. However, this algorithm directly projected image on the primary visual cortex, which ignored the process of the lateral geniculate, so we studied contour and boundary detection based on the combination of receptive filed, although this algorithm is superior to the algorithm based on the 2D-Gabor function, it doesn't consider the inhibition of non-classical receptive field, which results to a lot of texture edges.Secondly, because of the inhibition strength of non-classical receptive filed is related to the difference of orientation between non-classical and classical receptive filed, so we proposed contour and boundary detection based on the combination of receptive field and orientation modulation, the experiments show that this algorithm can effectively reduce texture edges.At last, since color information is very important for understanding the natural scene, we studied contour and boundary detection based on visual color theory, the color and brightness boundaries was detected by simulating color-opponent mechanism in visual path. However, the results had many texture edges, considering the homogeneous suppression of non-classical receptive filed, we proposed contour and boundary detection based on visual color theory and homogeneous suppression, the experiments show that this algorithm achieved good results in natural images.