Gender analiysis on iris images using novel machine learning techniques

Due to the recently increasing population, determining the gender of the person searched in the database means halving the number of data in the database, which will give us a great deal of time. In addition, the gender detection system can be used in security applications requiring gender-based access control, to organize marketing strategies that address only a specific gender group. It is shown that it is seen that the gender detection system has a wide application area. Many biometric features such as fingerprint, face, voice and signature are used in person identification systems. However, due to the unique structure of the iris, it is thought to be a more reliable system compared to other biometric properties. Therefore, in this study, it was tried to estimate the gender by using iris structure. A total of 3000 images were used by using right and left iris images of 750 women and 750 men. Texture analysis, geometric features extraction, Gabor filter, wavelet trasnform and fourier transform were applied to the iris images as features extraxtion techniques. In addition, general, regional and cellular texture analysis was performed on both the iris image and Daugman transformed images. Max normalization and non-linear transformation were applied to these extracted features for making scale invariand. Each feature set extracted each method is used to classify gender and compare echa others. It was clear that there is no best feature set of them. Therefore, after comparing all features echother according to accuracy feature selektion algorithm used to determine subset which had best classificaion accuracy. best individual selection, forward selection and n-take r-away selection algorithms are used as features selection algoritm. The selected features were compared each other. Consequently, the best feature subset was obtained nonlinear transform features with 91% accuracy rate using n-tak r away algorithm.