Optimal Pairwise Geometric Histograms


Abstract

This paper describes the application of a Multiobjective Genetic Algorithm (MOGA) to optimise the selection of parameters for an object recognition scheme: the Pairwise Geometric Histogram (PGH) paradigm. The overall result of the algorithm is to select PGH parameters giving a more compact, efficient histogram representation. The MOGA applied uses Pareto-ranking as a means of comparing individuals within a population over multiple objectives. The significance of this work is that it enables the process of pairwise object recognition to be fully automated so that it can reach the full potential for use on large databases. In future work these `optimal' histograms will be incorporated into the recognition process.