Abstract: The difference in the perception of light intensity variation between the human visual system (HVS) and commonly used linear and shift-invariant operators for edge detection is analyzed. Based on the principle of visual perception, a new model and an algorithm for detecting edges are proposed. And the edge detector obtained is both a locally and globally adaptive system for feature extraction. The results of experiments on a set of images of outdoor scenes containing tiny objects show that, compared with local operators, the operator proposed exhibits excellent capability of extracting edge features from images with low contrast.