Abstract: Templates of Kirsch operators are decomposed into difference templates and a common template. By a contrast between every two-neighbor difference templates, the direction of maximum edge intensity is found and the corresponding value of edge intensity is worked out. Thus it is no longer necessary to compute the edge intensity in eight directions, greatly reducing the convolution between templates of Kirsch operators and original image, and at the same time accelerating the convolution between the common template and original image by integral image that has been worked out in gray information processing. Using such a fast algorithm, Kirsch edge detection is made much less time-consuming than that of the current mainstream fast algorithm (FKC algorithm). The validity is proved by experiments. The idea of reducing convolution with templates decomposition and integral image has some universality. Examples show that this idea can also be applied in other edge detection algorithms and space filters.