Coupled feature selection for cross-sensor iris recognition

摘要: It is necessary to match heterogeneous iris images captured by different types of sensors with an increasing demand interoperable identity management systems. The significant differences among multiple such as optical lens and illumination wavelength determine the cross-sensor variations texture patterns. Therefore it a challenging problem select common feature set which effective for all sensors. This paper proposes novel optimization model coupled selection recognition. objective function our includes two parts: first part aims minimize misclassification errors; second designed achieve sparsity in spaces based on l2,1-norm regularization. In training stage, proposed can be formulated half-quadratic problem, where iterative algorithm developed obtain solution. Experimental results Notre Dame Cross Sensor Iris Database CASIA cross sensor database show that features selected method perform better than those conventional single-space methods Boosting h regularization methods.