Toward an optimal object-oriented image classification using SVM and MLLH approaches

Abstract

Support Vector Machines (SVM) and Maximum Likelihood (MLLH) are the most popular remote sensing image classification approaches. In the past, SVM and MLLH have been tested and evaluated only as pixel-based image classifiers. Moving from pixel-based analysis to object-based analysis, a fuzzy classification concept is used through eCognition software [1]. In this paper, SVM and MLLH are separately adopted and compared for multi-class object-oriented classification process where the input features vector contains primitive image objects produced by a multi-resolution segmentation algorithm. In this study, the determination of suitable object segmentation scale leading to an improved object-oriented classification result is also discussed and performed. Comparative analysis clearly revealed that higher overall classification accuracy (97%) was observed in the object-based classification using the optimal segmentation scale.