摘要：Boundary extraction is an important procedure associ-ated with recognition and interpretation tasks in dig-ital image processing and computer vision. Most ofthe segmentation techniques are based on the detec-tion of the local gradient, and then their applicationin noisy images is unstable and unreliable. There-fore global mechanisms are required, so that they canavoid falling into spurious solutions due to the noise.In this paper we present a gradient-based evolutionaryalgorithm as a heuristic mechanism to achieve bound-ary extraction in noisy digital images. Evolutionaryalgorithms explore the combinatory space of possiblesolutions by means of a process of selection of thebest solutions (generated by mutation and crossover),followed by the evaluation of the new solutions (fit-ness)andtheselection of a new set of solutions. Eachpossible solution is in our case a contour, whose fit-ness measures the variation of intensity accumulatedalong it. This process is repeated from a first approxi-mation of the solution (the initial population)eitheracertain number of generations or until some appropri-ate halting criterion is reached. The uniform explo-ration of the space of solutions and the local minimaavoidance induce to form better solutions through thegradual evolution of the populations
