A novel methodology for the interoperability evaluation of an iris segmentation algorithm

The performance of an iris recognition system depends greatly on how well the iris segmentation part of the system performs its task. The performance of an iris segmentation algorithm can be evaluated using different criteria and methods. Some of the methods evaluate the performance of the segmentation algorithm based on the performance of the whole iris recognition system. Other methods evaluate the performance of an iris segmentation subsystem independent of the performance of the system´s other subsystems. To our knowledge there do not exist a generally accepted method or criteria for the evaluation of the standalone iris segmentation subsystem. This paper proposes a novel methodology to compare the performance of different iris segmentation algorithms, applied to different image datasets in a consistent way. The methodology employs the F₁ score and an empirical cumulative distribution function. The implementation of the F₁ score estimation, adapted to the iris segmentation task is described. Finally the application of the proposed methodology is demonstrated and discussed.