Performance evaluation of change detection

This paper studies five different methods proposed for change detection thresholding. Methods were tested on synthetic and real pictures and results are quantitatively and qualitatively presented. Quantitative analysis is performed at pixel level and the following measures were used: the percentage of correct classification, the Jaccard coefficient and the Yule coefficient. We test the performance of thesholding methods in four experiments and Kapur was the best performing method both quantitatively and qualitatively.