Genetic Algorithm-Based Solver for Jigsaw Puzzles - Analysis and Improvement

An analysis of the GA-based jigsaw puzzle solver was performed. Reproduction stage crossover operator proves to be the core part of the algorithm, using the best buddy property for fast solution convergence. Number and validity of best buddy pieces depend on the compatibility metric used. LPQ compatibility metric provides the best result, achieving improvement of 5% with respect to the used SSD metric. A crossover modification is proposed resulting in 8% increase in average accuracy.