A Multiresolution-Fuzzy Method for Robust Threshold Selection in Image Segmentation
José Ramiro Martínez‐de Dios, Anı́bal Ollero · Intelligent Automation & Soft Computing · 2006
Abstract Threshold selection represents an essential topic in computer vision. The degradation of classical threshold selection methods when dealing with images with changing conditions limits the utilization of these methods in vision applications in unstructured environments. This paper presents a threshold selection method that can be easy adapted to the pazticularities of a vision application. The method analyses coarse-to-fine approximation descriptions of the histograms of the image. At each level of resolution the histograms descriptions are decomposed in histogram modes, some of which aze classified as object modes by a fuzzy supervising system designed to adapt the analysis to the particularities of the application. The method shows high robustness to changes in illumination conditions and image noise. Some examples and its utilization in two applications aze presented.