A noise-smoothing nonlinear operator for the preprocessing of text and graphics images

When trying to enhance the quality of a gray-level digital image obtained from the acquisition of a document containing text and/or graphics, relevant problems arise if the original is not highly contrasted and if it is spoiled by noise. In the paper, a nonlinear preprocessing method is suggested for this application. It is based on a two-dimensional (2-D) complete quadratic filter, i.e. an operator composed by a linear FIR component and a quadratic component acting, in our case, on the same support. In the proposed operator, the linear component is a conventional low-pass filter having a noise-smoothing effect; the quadratic component, on the other side, being sensitive to the abrupt changes in the luminance levels of the input image, compensates for the blurring effect of the former term and is even able to enhance the edges of the objects. The resulting image is thus less noisy and sharper than the original one.<
>