A technique, specifically apparatus and an accompanying method, for accurately thresholding an image based on local image properties, specifically luminance variations, and particularly such a technique that uses multi-windowing for providing enhanced immunity to image noise and lessened boundary artifacts. Specifically, a localized intensity gradient, G(i,j), is determined for a pre-defined window (300) centered about each image pixel (i, j ) . Localized minimum and maximum pixel intensity measures, L.sub.min and L.sub.max, respectively, are also determined for another, though larger, window (330) centered about pixel (i,j). Also, a localized area gradient measure, GS(i,j), is determined as a sum of individual intensity gradients for a matrix of pixel positions (370) centered about pixel position (i,j). Each image pixel (i,j) is then classified as being an object pixel, i.e., black, or a background pixel, i.e., white, based upon its area gradient, GS (i,j), and associated L.sub.min and L.sub.max measures.