Affine object location is a difficult problem in computer vision. Genetic algorithm (GA) provides an efficient solution to the problem when there is little noise or other artifacts. Nonetheless, there is a need to design better GA for complicated, noisy images. In this paper, we investigate the performance of improved operators in GA designs to solve the above problem. Three operators are investigated: redundancy checking, adaptive mutation and partial reshuffling. An intuitive discussion of why these operators are helpful is given in terms of the basic working principle of the GA. Experimental results on the probability of success are given using the framework of Repeated Genetic Algorithm (RGA), which is an application of the probabilistic amplification technique. Tests on both synthetic and real images, with random and structured noise are conducted. It is found that whilst all three operators can improve the GA’s probability of success for the affine object location problem, the redundancy checking operator is the most effective.