Combining a fuzzy classifier with classifiers based on statistic moments

摘要

This paper describes the behaviour of a global classifier constituted by a fuzzy classifier and two distance classifiers based on statistic moments. when we try to develop a classifier to recognise with successful a set of objects, we must have in mind that is impossible to obtain perfect performances (100% of recognition). However, performances between 80% and 95% are normally obtained with only one classifier. This fact, derive from the type and number of features used, which sometimes isinsufficient to eliminate faults in the classification. It was developed a global classifier constituted by a combination of three classifiers based on two types of features. One is based on the visual perception such as area, perimeter and index ofcompactness, and the other two are based on statistic moments. The combination of the three classifiers is done using the voting principle.The study was applied on binary images in a set of objects with various positions and orientations on the image plan. The study showsthat is possible, with good results, the combination of the flizzy set theory with statistic theory on a multipleclassifier.