Skin Lesion Segmentation Based on Region-Edge Markov Random Field

Abstract

This paper presents a probabilistic model based on Markov Random Field (MRF) theory to achieve skin lesion segmentation. MRF theory plays a significant potential role in the image segmentation field. It has several models based on its theory such as region-based MRF model and edge-based MRF model to detect object, boundaries and other relevant information in an image. The proposed method aims to combine the advantages of these two models by computing the product of the regional likelihood function and edge likelihood function. Regional features and edge features are used to solve the maximum a posteriori (MAP) estimation problem to find the best estimation for better image segmentation. The algorithm starts from pre-processing obtained by convolution technique, and iteratively refines the segmentation by taking into account several metrics of region homogeneity under a probabilistic framework. The technical content is described in detail, and the algorithm was tested on the International Skin Imaging Collaboration (ISIC) database, showing its potential. The proposed method shows a significant improvement when compared with individual lesion segmentation methods in ISIC 2018 challenge with overall results achieved as Jaccard Index of \(76.40\%\).