RGB Image Segmentation Using Hierarchical Analysis of 2D-Histograms. Application to Menthe Plant

In this study, segmentation of RGB images into two classes (soil and vegetation) is performed by unsupervised classification method based on the hierarchical analysis of 2D-histograms. The 2-D histogram is considered as a monochrome image. The approach we develop in the present paper is based on the classical waterfilling algorithm, which we will apply to monochrome image. We perform the segmentation algorithm by reclassification of the pixels not classified in the determined classes based on a colour distance. The Mahalanobis distance gives best results than the Euclidean one. The proposed approach has been tested on synthetic and real RGB colour images and the results are satisfactory. For the menthe plant image, two classes are obtained by taking a threshold between 8% and 18%, which gives much latitude in the choice of S