Applications that can be formulated as Open-set Identification or Verification tasks are often vulnerable to noisy data (e.g., tone of a recorded voice command). Score normalization techniques effectively address this problem by mapping match scores between the gallery and probe samples to a common domain. The Rank-based Score Normalization Framework is a method that can treat multisample galleries more efficiently. This invention does not require tuning of any parameters and it can be used in conjunction with any score normalization technique and any combination rule. Specifically, the framework consists of three algorithms where each of them is novel in a different manner: (i) Algorithm 1 partitions scores and normalizes each resulting subset independently, (ii) Alg. 2 uses information from the gallery versus gallery similarity scores in conjunction with Alg. 1, and (iii) Alg. 3 updates the information in an online fashion. Results using Face Data indicate significant improvements in terms of recognition performance.