Study the Effect of Convolutional Local Information-Based Fuzzy c-Means Classifiers with Different Distance Measures

Abstract

For improving image attributes and increase robustness to handle the noise for image classification, fuzzy algorithms were introduced. The fuzzy c-means (FCM) algorithm has not been suggested to incorporate spectral and spatial neighborhood information for accomplishing image classification. In this study, evolution developed for overall accuracy (OA) calculated by FERM (fuzzy error matrix) technique for an algorithm, namely fuzzy local information c-means (FLICM) using base classifier FCM, adaptive fuzzy local information c-means (ADFLICM) using base classifier FCM and fuzzy c-means with constraints. For these algorithms, various distance measures and weighting components (m) are studied. This study provides a comparison among different forms of local convolution with an FCM classifier, with a combination of different distance measures and weighting components to select the best classification approach. FLICM using base classifier FCM and Euclidean distance measures with a weighting component (m) of 1.3 has resulted in the highest overall accuracy compared to other classifiers used with various distance measures.