Algorithms of background suppression in the problem of detection of point targets in images

Abstract

Various approaches to estimation and suppression of a motionless background with the use of texture correlations in the problem of detection of small-size dynamic targets are considered. Algorithms of suppression of a locally flat background, background suppression by means of bilateral filtration, and an algorithm of background estimation and suppression with the use of an autocorrelation function are implemented. For anisotropic textures with boundary transitions, an algorithm of background estimation and suppression along the boundary and an algorithm of three-channel filtration are proposed and implemented. Operation of these algorithms on textures representing different classes of images is compared. It is demonstrated that the algorithm with background estimation along the boundaries yields good results for model data with a large number of linear boundaries, but its operation on mixed-type textures is less efficient than that of other available approaches. Among the considered algorithms, the approach based on three-channel filtration ensures the greatest increase in the signal/noise ratio for various textures modeling real images.