Boundary Extraction Through Gradient-Based Evolutionary Algorithm

Boundary extraction is an important procedure associated with recognition and interpretation tasks in digital image processing and computer vision. Most of the segmentation techniques are based on the detection of the local gradient, and then their application in noisy images is unstable and unreliable. Therefore global mechanisms are required, so that they can avoid falling into spurious solutions due to the noise. In this paper we present a gradient-based evolutionary algorithm as a heuristic mechanism to achieve boundary extraction in noisy digital images. Evolutionary algorithms explore the combinatory space of possible solutions by means of a process of selection of the best solutions (generated by mutation and crossover), followed by the evaluation of the new solutions (fitness) and the selection of a new set of solutions. Each possible solution is in our case a contour, whose fitness measures the variation of intensity accumulated along it. This process is repeated from a first approximation of the solution (the initial population) either a certain number of generations or until some appropriate halting criterion is reached. The uniform exploration of the space of solutions and the local minima avoidance induce to form better solutions through the gradual evolution of the populations.