Automated detection of subtle changes in sequential images requires development of procedures that ensure true subtraction of two images by proper registration and normalization of illumination. The difference image can then be further processed to analyze the spectral components and the gray level variations. An innovative and generalized approach to accurate image subtraction incorporating illumination normalization and correction for translation, rotation, and scale difference in two sequential images is considered. The entire procedure yielding true image subtraction has been implemented in a PC-based image processing system and can be executed by menu-driven operations. The effectiveness of this approach is illustrated through subtraction of registered sequential fundus images and analysis of the spectral and gray-level variations of the difference images.