A Generic Methodology for Image Segmentation Evaluation

In this paper, we present a new methodology to evaluate image segmentation algorithms in a supervised context. The suggested measure is built according to defined quality criteria as shape parameters and homogeneity criterion between regions. This new methodology has been tested to evaluate some usual segmentation algorithms by comparing the resulting segmented image against a manually segmented reference image (ground truth). The results have been compared to those given by existing measures. Experimental results show that the suggested approach performs better than existing methods