Efficiency of spherical filters on detection of isotropic defects in textured backgrounds

Abstract

This paper concerns the detection of small defects inserted in various textured backgrounds with more or less spherical filters and wavelets. We have evaluated the detection efficiency of the filters when the controlled isotropic defects are first added in synthetic textured images, then in real reference textured images, the Brodatz textures, and finally in medical images, parts of digital mammographies. Three families of filters are involved: the less spherical family is the Gabor filters, the nearly isotropic wavelets φ and ψ, and the Mexican hat filters, which are totally spherical. We have also studied the influence of the defect amplitude by considering various truncations. To achieve this, the defect height was truncated at different percentages.