A feature-level solution to off-angle iris recognition
Xingguang Li, Libin Wang, Zhenan Sun, Tieniu Tan · 2013
For iris recognition, it is inevitable to encounter a large portion of off-angle iris images in less constrained conditions. This paper proposes a feature-level solution to off-angle iris recognition which is less dependent on iris image preprocessing. Firstly, we use geometric features of corneal reflections and multiclass SVM to classify iris images into five categories (i.e., frontal, right, left, up and down) according to the off-angle orientation of iris region. And then a feature learning method based on linear programming is used to select the most effective ordinal features of each iris category. Finally, the input off-angle iris image is recognized with the specific ordinal feature template belonging to the corresponding iris category. Experimental results on the Clarkson Angle database demonstrate that our feature-level solution significantly outperforms the mainstream methods based on off-angle iris image preprocessing.