A neural network based approach for iris recognition based on both eyes

A novel approach for iris recognition using both eyes is implemented and tested. The right and left eyes of three individuals from the CASIA Iris database V3 are used. The iris is extracted from the whole eye, normalized and enhanced. A one level discrete wavelet transform is applied on the enhanced iris sheet. The mean of an 8×8 or 4×4 blocks of the approximation coefficients is evaluated and the iris code is generated. A system trained using the right eye code and tested using the left eye code achieved a max recognition rate of 75%. The system is then trained using half the data from the right eye and the other half from the left eye. The rest of the data is used for testing. When using the approximation coefficients, a vector of length 1647 is obtained and a recognition rate of 98.3% is achieved. The mean of 8×8 blocks of 1st approximation coefficients is then used in the same manner resulting in reduction in the code size while improving the recognition rate which reached 100% in this case. The same was done for the approximation of a 2nd level discrete wavelet transform where the mean of 4×4 blocks is calculated and employed to generate a feature vector of a much reduced size. The recognition rate reached also 100% in this case which verifies the superiority of the suggested technique.