Optimal algorithms for target location in non-homogeneous binary images.

Binary images appear in various pattern recognition applications. It is thus important to design pattern recognition algorithms that are optimal for such images. We address the problem of target location in binary images perturbed with nonhomogeneous background noise. The proposed algorithm optimizes the likelihood ratio between the hypotheses that the target is present within a small subwindow of the image and that it is not present in this subwindow. The algorithm is shown to consist of two correlation operations and a few pointwise nonlinear transformations. With numerical simulations, we illustrate the efficiency of this technique, especially in the presence of strongly nonhomogeneous background noise