Score level fusion of SIFT and SURF for iris

Iris detection depending on local features like SIFT (Scale Invariant Feature Transform) and SURF (Speeded up Robust Features) exhibits high accuracy though the approaches leave behind further scope for improvement due to the lack of proper choice of score generating function and score fusion. Usually the score of a matching algorithm is taken to be number of matches. However a properly chosen function of number of matches can also be considered as a score. The proposed approach in this paper performs a classification operation on the detected keypoints. Each set of the keypoints of the query image is subjected to nearest neighbour match with respective set of keypoints of the database image. Hence there are two scores generated by the matching of two classes. This paper also proposes a mathematical monotonic function on these two scores to generate a single score such that the final score value gives rise to better disjunction between genuine and imposter scores than conventional SIFT.