Multi-source Heterogeneous Iris Recognition Using Locality Preserving Projection

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

“…To evaluate the contribution of our proposed algorithm, we compare it with related algorithms. One of these utilises unsupervised LPP [19], and the other utilises SRC [9]. In [9], Gabor features are extracted from the iris region.…”

“…In [19], Gabor features are extracted from iris regions. Dimensionality reduction is performed using unsupervised LPP.…”

“…As expected, increasing the number of features improves the recognition rate to a certain extent, while recognition time generally increases. Our algorithm (SLPP + SRC) gives the highest recognition rate, of about 99.5% at 100 features, while taking the lowest time of about 5 ms. Unsupervised LPP with SVM (LPP + SVM) [19] gives a significantly lower recognition rate, 84.2% at 120 features, while taking 30 ms. Finally, the random projections with sector‐based SRC (Rand + SSRC) [9] method gives a reconstruction rate of 93% at 120 features while taking 410 ms.…”

“…Other works employ LPPs for dimensionality reduction. For instance, [19] utilised LPP to perform manifold learning. Subsequently, an SVM classifier is utilised to find and establish the optimal classification hyperplane in low-dimensional space.…”

“…Alternatively, many works extract Gabor features from iris images and then apply dimensionality reduction techniques to the extracted features. For instance, [19] applies LPP to extracted Gabor features at five scales and eight orientations. Some works introduce other features for use in iris recognition.…”