MAX-MIN Ant System in Image Preprocessing

Abstract

MAX–MIN Ant System (MMAS) application in image preprocessing is investigated. Standard MMAS model for traveling salesman problem is presented together with MMAS model modifications, applying it for image preprocessing. Two modifications of initial ant placement strategy introduced, one based on simplified MMAS without heuristic information (Mod1), second is based on normalized quantity of moved ants (Mod2). Experimentally determined percentage of moved ants is 20%. Provided modifications were tested on synthetic images with evaluation of convergence speed. Additionally test results were compared with simple brute force solution finding method. Initial ants placement based on pheromone concentration proved to be an effective way to increase convergence speed. With solution length of 6 operators 30% increase in convergence speed was achieved compared to MMAS without pheromone control. Mod2 showed 7% decrease in quality on short run problems (5 operators), however on longer solution (6 operators) Mod2 solution quality decrease slope was less rapid (quality decrease 25%) compared to both standard MMAS without initial ant placement strategy and Mod1 (quality decrease 49%). Analyzing deviation of number of iterations Mod2 also showed less rapid increase in deviation compared to MMAS and Mod1. Ill. 2, bibl. 11 (in English; summaries in English, Russian and Lithuanian).