Discriminative masking for non-cooperative iriscode recognition

The reliability of iris-based biometric recognition faces challenges due to non-cooperative image acquisition. This paper presents a novel non-cooperative recognition method by adopting a step-by-step segmentation process of the iris regions and modifying the generated IrisCodes in a way that the bits having significant discriminatory power are selected using an effective masking process. Both the statistical inter- and intraclass consistencies are used to estimate the discriminatory power of each bit position of the IrisCodes. Experimental results on commonly-referred non-cooperative iris image database show the significance of using discriminatory bits to improve the recognition accuracy for both the log-Gabor and discrete cosine transform-based IrisCodes.