Editorial : Neural - Fuzzy Applications in Computer Vision

Neural networks and fuzzy logic are two bio-mimetic techniques that are used to provide approximations to real-world problems. While for some people this biological plausibility provides some justification for their use, for others the important point is that, regardless of their origins, both approaches are known to be robust alternatives to conventional deterministic and programmed models. However, the two paradigms have distinct application domains. Fuzzy logic is used to represent qualitative knowledge, and provides interpretability to system models. By this we mean that a system model is explicit and is understandable to a knowledge or systems engineer. This facilitates inspection of the model, and therefore simplifies and encourages its validation and maintenance. Zadeh [14] has summarised fuzzy logic as a body of concepts and techniques for dealing with imprecision, information granulation, approximate reasoning and computing with words. Neural networks, on the other hand, are used to induce knowledge or functional relationships from instances of sampled data. This is useful when it is not possible to develop analytic models from first principles but the system is observable. However, in contrast to fuzzy systems, this knowledge is not readily understandable to the system designer because it is encapsulated in the so-called black box. Another contrasting feature of neural and fuzzy techniques is that while traditionally fuzzy knowledge is obtained from human experts, neural network relationships are usually automatically learned from a training process that iterates through a sample data. In consideration of this, the combination of fuzzy and neural systems provides a synergy such that the marriage of each of their strengths overcomes some of their individual drawbacks and can lead to greatly enhanced systems. In particular, fuzzy system design does not incorporate any learning, while neural networks do not possess mechanisms for explicit knowledge representation. Fuzzy processing is desirable in computer vision because of the uncertainties that exist in many aspects of image processing. These uncertainties include additive and non-additive noise in low-level image processing, imprecision in the assumptions underlying the algorithms, and ambiguities in interpretation during high-level image processing. For example, for computational convenience, the common process of edge detection usually models edges as intensity ridges. Nevertheless, in practise this assumption only holds approximately, leading to some of the de