Perspectives on fuzzy systems in computer vision

The problem of computer vision is to automatically characterize the contents of digitized images. Applications include factory automation, navigation, digital libraries, and medicine. Not only is recognition an “inverse problem” with no single mathematical solution, but it is also complicated by external sources of uncertainty such as the conditions of image formation. Thus, the need for dealing with uncertainty in computer vision is well accepted. However, the majority of work in this area has used fixed thresholds or probabilistic approaches, from surface reconstruction to object recognition. The paper surveys current approaches to uncertainty in computer vision, paying particular attention to the attitudes toward fuzzy systems. Although fuzzy systems are out of the mainstream of computer vision, they pose great promise for addressing uncertainty issues that are not adequately dealt with by current methods