How robust is the SVM wound segmentation?

Abstract

This paper investigates the robustness of automatic wound segmentation. The work builds upon an automatic segmentation procedure by the support vector machine (SVM)-classifier presented in [M. Kolesnik et al. (2004), (2005)]. Here we extend the procedure by incorporating textural features and the deformable snake adjustment to refine SVM-generated wound boundary. The robustness of SVM-based segmentation is tested against different feature spaces using a long sample of training images featuring a broad variety of wounds' appearance. Recommendations drawn from these experiments provide a useful guideline for the development of a software support system for the visual monitoring of chronic wounds in wound care units