Clutter can be a significant challenge to the detection, tracking and discrimination abilities of infrared seekers and sensors, creating considerable interest in techniques for its characterization and synthesis. In the past, clutter power spectral density models have been used both to analyze and generate infrared clutter images. Many of these models assume stationary statistics for the clutter process (i.e., the clutter is homogeneous). For many real images, however, nonstationary statistics are needed for characterization. Likewise, more realistic synthetic images can be generated through the use of nonstationary statistics. This paper presents a technique for segmentation of an image into homogeneous regions and for subsequent synthesis of a similar image from white noise. For the segmentation, two features, the slope of the local-area power spectrum and the local-area mean, were first used to characterize each pixel in the image. Then, using the feature values, a 2D histogram was formed and a Bayesian decision process was used to cluster pixels with similar features into a small set of classes. The synthesis technique uses an adaptive spatial- domain filter to generate clutter images from white noise.