fielding an instrument which will be capable of classifying acoustic signatures under battlefield conditions in as near to a real-time manner as possible. This burden places severe restrictions on the complexity and robustness of the system that is ultimately fielded.
Our second application focuses on the automatic segmentation of images into regions of homogeneous content, with an ultimate goal of classifying the different regions. Our research has focused on a wide variety of image modalities and desired classification systems. They include the detection of combat vehicles in natural terrain (Solka, Priebe, and Rogers, 1992), the detection of man-made regions in aerial images (Priebe, Solka, and Rogers, 1993 and Hayes et al., 1995), and the characterization of mammographic parenchymal patterns (Priebe et al., 1995). In Figure 2, we provide representative vehicular and mammographic images. The homogeneity criterion in this case is based on functionality of the region. For example, we wish to