Recognition of multidimensional affine patterns using a constrained GA

The problem of determining an affine relation among multidimensional data points is addressed in this paper. In the first step of the illustrated procedure, the parameters for the linear manifold that fits the data are determined in closed form using a (weighted) total least squares formulation of the problem. The solution obtained, however, is highly sensitive to data points with exceptionally high noise (outliers). The problem of outliers suppression is then formulated as a constrained binary optimization problem and a genetic algorithm with nonstationary penalty function is used to solve it efficiently