A scale selection mechanism of based on physiology

This paper studies the scale we choose for boundary detection. The common multi-scale method for detecting boundary requires the users to compute several scales of boundary on the whole image, train the coefficients as the weight of each scale's information, and combine the multi-scale boundary using the training weight. It has been proposed that human visual system evolved to be able to change the size of its receptive fields according to the object, which gives us inspiration to design physiologically scale varying boundary detection models. In this paper, we propose an automatic scale selection mechanism for each pixel according to its “cell activity”. Then we run the boundary detection algorithm using the optimal scales that the mechanism selects.