In this paper, we present a content-based filtering technique to enhance scanned documents. An image classification step is performed to classify each pixel into text, background, and image regions. With the segmentation step, we can strongly sharpen text and similar edge detail while smoothing background and image content. To optimally select the segmentation parameters, we formulate a cost function to minimize the number of miss-classified pixels between classified test and reference images. This cost function is minimized using genetic algorithms.