Global pattern classification in dermoscopic images based on modelling

Abstract

In this paper a model-based method of classificationof global patterns in dermoscopic images is proposed. Global patterns identification is included in the pattern analysisframework, the melanoma diagnosis method most used amongdermatologists. The modelling is performed in two senses: first adermoscopic image is modelled by a finite symmetric conditionalMarkov model applied to colour space and the estimatedparameters of this model are treated as features. In turn, thedistribution of these features are supposed that follow a Gaussian mixturemodel along a lesion. The classification is carried out by an image retrieval approachwith different distance metrics. A 78.44% success rate in average is achieved when globular, homogeneous, and reticularare classified and a 72.91% success rate when the multicomponent pattern is added.