A new iris recognition algorithm

based on PCA and ICA

is proposed in this paper. Firstly

PCA was applied to the iris images in order to reduce dimension and second order correlation

then ICA was applied to train iris images. In our algorithm

ICA was performed on iris images in the CASIA database under two different architectures

of which one treated the image as random variables and the pixels as outcomes

while the other treated the pixels as random variables and the images as outcomes. The first architecture found spatially independent basis images for the iris. The second architecture used ICA to find a representation in which the coefficients used to code images were statistically independent. No matter which architecture we used to train the iris images

the proposed algorithm was effective.