Filtering feature mismatches using genetic algorithm and fundamental matrix

We propose an efficient method to filter the incorrect matches of SIFT-based feature matching which includes correct and incorrect matches using fundamental matrix and genetic algorithm. First, best fundamental matrix is estimated by genetic algorithm whose input data are unfiltered feature matches. To design chromosome, each gene represents the index of selected matches and it has linkage value which is got from linkage map. Selected matches are used to estimate fundamental matrix candidates. Each chromosome has a fitness value related to the number of found correct matches and matching score which is calculated from its fundamental matrix. To crossover, multiple order crossover operation is applied. For replacement, k-rank chromosomes are replaced from sorted global generation. Then, best fundamental matrix is used for filtering. Finally, we show that good matches are extracted comparing RANSAC and LMED fundamental matrix estimation.