Efficient high‐speed framework for sparse representation‐based iris recognition

Abstract

While various frameworks for iris recognition have been proposed, most lack efficiency and high speed. A new framework for iris recognition is presented that is both efficient and fast. Feature extraction is performed by extracting Gabor features and then applying supervised locality‐preserving projections with heat kernel weights, which improves the recognition rate in comparison with the results from unsupervised dimensionality reduction techniques such as principal component analysis, locality‐preserving projections, and random projections. Afterwards, a classification is performed using the recently proposed sparse representation‐based classification (SRC). To considerably improve classification performance, SRC is proposed, using a greedy compressed‐sensing recovery algorithm, as opposed to employing the traditional computationally expensive ℓ1 minimisation. The proposed framework achieves a recognition rate of about 99.5% using two iris databases, with a significant improvement in speed over related frameworks.