Characterization and Numerical Evaluation of a Color Image Segmentation Method

Abstract

We present a characterization and a numerical evaluation of an own semi-automatic color image segmentation method using generated synthetic images with its associated ground truth. The evaluation methodology was designed to assess the efficiency of the resulting color information achieved from color segmentation algorithms. By the use of ROC curves and its analysis, we obtained some particular characteristics of our segmentation method, such as the level of sensibility related to the threshold selection and to the appropriate number of pixels to have by the color sample needed by the algorithm.