Adaptive Fuzzy Inference for Edge Detection Using Compander Functions And Linear Fitness Function Transformations

摘要

As an initial step in image processing, edge detection is one of the most important operations which must provide a good separation between an object and its background. This paper addresses the edge detection problem by using a combined approach of several techniques including multi-resolution and fuzzy reasoning. The techniques used here are based upon the Canny edge detection algorithm but uses a fuzzy reasoning generated from an adaptive neural network based fuzzy inference system to select the resolution and threshold parameters. As an alternative to the classical fuzzification process, the A-Compander method is shown to work equally as well and has the appealing advantage of selecting the fuzzy inference parameters in a compact manner. Further, to improve learning during the tuning phase, a linear transformation of the fitness function is employed. An example image is investigated to illustrate the approach and results are comparable to the classical adaptive fuzzy neural inference and the multi-resolution approaches, but with less computational requirements.