Edge Detection with Self-Adaptive MAS Using Multi-distance Gradient

This study proposes a novel application of MAS (Multi-agent System) framework for edge detection, which the edge definition is derived from the compositive behavior of the gradients of different sample distance. The edge feature can be regarded as an extension of traditional gradient-based methods by the product of gradient magnitude on different distance. And then MAS is employed to obtain the maximal edge response for MAS having parallel operation while the individual agent being simple and easy to implement. Thus it makes the method adapt to strong or weak edge because of adopting the local extremum instead of global one. Furthermore, some strategies are appended for searching the edge, included shifting and mutation of the individual agent. With the self-increasing threshold, the simple agents reach the ideal edges according to the mutual cooperation. Finally, experiments prove that edge detection result of the proposed method is better than other intelligent ones and traditional ones, besides, it is not sensitive to the noise.