Unsupervised Color Image Segmentation Based on Non Parametric Clustering

Abstract

Many segmentation problems have been addressed using probabilistic modeling. These methods tend to estimate the region membership probabilities for each pixel of the image. The segmentation results depend strongly on the initialization of these regions and the selection of the appropriate number of segments. In this paper we present an unsupervised segmentation method based on non parametric clustering able to deal with these two issues. After a simple splitting, a minimum variance criterion is used to generate both the initial regions and their number. The proposed model was applied on various images (synthetic, natural) showing good visual results. Finally numerical experiments demonstrate the efficiency and the robustness of the proposed model compared to other segmentation methods.