The Contour Detection Research Based On Biological Vision Theory

Contour detection has become a critical step for high-level visual tasks.Sufficient physiological studies show the existence of a mechanism,called surround suppression,which can contribute to contour detection.Although the previous works have proposed many contour detection models,it is still a challenge to detect contours with high precision and efficiency in complex images.There are currently two difficulties: one is missing contours,and the other is hard to remove texture edges.For the two issues,this paper proposes two modified contour detection algorithms based on biological vision mechanism.The contours detected by the traditional surround suppression algorithm still have a large number of redundant texture edges,and the detected contours are not complete yet.Since superpixel boundaries could adhere well to image contours,We utilize a modified superpixel generation process to generate a set of contour candidates,and then take the pixels on the superpixel boundary which occur multiple times as candidates,this can reduce the effects of noise and texture on the contours.Further,redundant texture edges in candidate contours are inhibited by an orientation-based surround inhibition model,and the contours are preserved.By comparing with multiple-cue-based models and other existing biologically-inspired methods,the experimental results show that the proposed algorithm has better performance.It is a contradiction to preserve the contours while suppressing the texture.This paper proposes a modified contour detection algorithm based on the multiple-cue-based surround suppression model.The weights are calculated using local features such as luminance,luminance contrast and orientation,and the weights of different cues are combined by a scale-guided strategy.The combined weights are then used to modulate the intensity of surround suppression.In this paper,the surround suppression strength is increased to further suppress redundant texture.And then an improved superpixel segmentation algorithm(SLIC)is used to obtain contour candidates,further,de-suppression is performed at the contour candidates.The experimental results show that the proposed algorithm suppresses redundant textures as much as possible while extracting more complete contours.The experimental results on the benchmark dataset show that the proposed methods have practical value.