This paper deals with performance evaluation of raster-to-vector conversion algorithms. We briefly review the past work, and focus our attention on a method proposed by Phillips and Chhabra \cite{phillips99a}. We show that the matching algorithm use d in their method can be improved by introducing new metrics and reformulating the problem as a combinatorial optimization scheme. Improvements brought by our approach are the reduction of the number of utilized thresholds to one, and a gain of stability in the obtained results. We also propose an algorithm to solve the scheme, and compare the evaluation results obtained with both methods on real CAD drawings.