System and method for sensor adaptation in iris biometrics

摘要

The sensor adaptation technique applicable to non-contact biometric authentication, specifically in iris recognition, is designed to handle the sensor mismatch problem which occurs when enrollment iris samples and test iris samples are acquired with different sensors. The present system and method are capable of adapting iris data collected from one sensor to another sensor by transforming the iris samples in a fashion bringing the samples belonging to the same person closer than those samples belonging to different persons, irrespective of the sensor acquiring the samples. The sensor adaptation technique is easily incorporable into existing iris recognition systems and uses the training iris samples acquired with different sensors for learning adaptation parameters and subsequently applying the adaptation parameters for sensor adaptation during verification stage to significantly improve the recognition system performance.

主权项

1. A method for iris recognition, comprising the steps of:
establishing a data processing sub-system in an iris recognition system configured to perform iris recognition through the steps of: performing a training stage of operation in said iris recognition system, including the steps of: (a) acquiring a plurality of training iris samples ={θ1, θ2, . . . , θN} with a plurality of different sensors {S1, S2, . . . , SNs} for a plurality of subjects {y1, y2, . . . yNc}, (b) for each pair θi,θj of training iris samples from said plurality thereof, computing, by said data processing sub-system of said iris recognition system, a similarity measure (θi,θj) between said training iris samples θi and θj, where θi, θjε, (c) forming, by said data processing sub-system, an initial training kernel matrix 0=(θi,θj) for said plurality of training iris samples, (d) iteratively updating, by said data processing sub-system, said initial training kernel matrix 0 by applying Bregman projections at each iteration, thereby forming a final training kernel matrix A(θi,θj), and thereby establishing a space of allowable transformations for said training iris samples θi,θj satisfying predetermined constraints {θi,θj} imposed between said training iris samples θi and θj, (e) extracting adaptation parameters σij through processing said initial and final training kernel matrices by said data processing sub-system; performing a testing stage of operation subsequently to said training stage of operation through the steps of: (f) acquiring a test iris sample θt, (g) computing, by said data processing sub-system, said similarity measure (θt,θ) for said testing iris sample θt and said plurality of the training iris samples θε, (h) forming, by said data processing sub-system, an initial test matrix 0test=(θt,θ), (i) computing, by said data processing sub-system, a test adapted kernel function Atest for said test iris sample θt based on said initial test matrix 0test and said adaptation parameters σij asA⁢(θt,θ)=⁢(θt,θ)+∑ij⁢σij⁢⁢(θt,θi)⁢⁢(θj,θ);  and performing a matching routine subsequently to said testing stage of operation through the steps of: (j) computing, by said data processing sub-system, a distance measure between said test iris sample θt and an enrollment iris sample θen, and (k) verifying, by said data processing sub-system, said test iris sample as genuine if said distance measure in said transformed space between said test and enrollment iris samples is smaller than a predetermined threshold.