This paper proposes a novel application of ant colony system for edge extraction which the food definition is combined the idea of the gradient and the multi-scale. After building the food map, ant colony system is employed to obtain the maximal food response, which makes the method adapt to strong or weak edge basing on the local extremum instead of global one. With the feedback of the pheromone and prior directional searching, the simple ants achieve ideal edges based on the direct response and the mutual cooperation. Furthermore, the method can work well in both gray level images and color images. Finally, the experiments prove that edge extraction result of the proposed method is better than other intelligent ones and traditional ones, and it is not sensitive to the noise.