Fuzzy Expert System for Object Labeling Integrated with Probabilistic Learning

This paper presents a fuzzy expert system that demonstrates a labeling method for multiple-object images based on the histogram analysis. The images are rock textures, and objects to be labeled are the constituent minerals. The system is used to explain the decision process of minerals identifications. The system introduces an approach that can be generalized to any domain-dependent images. The approach applies a probabilistic unsupervised learning on the image, which then provides presentation for the image with fewer gray levels. The histograms of the clustered images are analyzed for turning points range estimation for the image set. The turning points indicate a transition from one object to another. The system correctly labels the four image categories namely Pores Space, Grinded Materials, Quartz, and Feldspar. The system is tested using 51 rock textures images. The system creates the corresponding segmented and labeled binary images of the image set