As the optimal operator of edge detection,classical Canny operator has the limits in linear filtering and threshold choosing.On the basis of it,an adaptive Gaussian filter with scale adjustable was adopted in filtering.By computing the average of all intensity values in current filter window and the difference between the average and the intensity value of the current pixel,the smooth degree of current filter window was obtained.And then,by regarding the smooth degree as the scale size of Gaussian function,the noises of images could be filtered while the edges couldn’t be smoothed.At the same time,an adaptive double-threshold selection method was used based on the grads threshold method and simple statistic method to detect edges of Non-Maxima Suppressed grads image.Experiments prove that the method can filter noise effectively;also the distilled fake edges and lost edges are fewer than those of classical Canny operator.