Ant colony optimization variants in image edge detection

Edges in an image are the curves consisting of pixels wherein both side contains pixels with non-uniform intensity. Edge detection is a part of low level image processing, much needed in various fields. Though edge detection can be done by various derivative techniques but it can also be detected well using meta-heuristic approximation algorithms. Ant Colony Optimization (ACO) is such a meta-heuristic technique to solve it. In basic ACO which comprises five phases: Initialization, Construction, Updation, Decision and Visualization, we have proposed and implemented total eight variations in this paper by modifying initialization and construction phase. In the initialization phase we have given a constraint in one variant that ants will be initialized near to edge to eliminate useless construction steps and unwanted edge detection where the other variant is without this constraint which may generate unnecessary edges in the resulting image. We have taken other two variations in selecting the next pixel in the construction phase: in one Greedy method is used, in another Roulette wheel selection method is used. Apart from these, in this phase two more variations have been done depending on memory size of ants i.e. applying tabu list memory of ants and ants without memory. Hence on the basis of two types of selection method used, two types of memory size of ants and two types of initialization phase, we have implemented eight variations individually in this paper. We observe that the variant, with roulette wheel selection, incorporated with the tabu list memory of ants, and with the new initialization condition outperforms others.