To search, Click below search items.


All Published Papers Search Service


Illumination Normalization for Fingerprint Recognition using Enhanced Multi-scale Low Rank + Sparse Decomposition with Artificial Neural Network


Omeed Kamal Khorsheed


Vol. 17  No. 6  pp. 8-13


Image recognition are very sensitive to light conditions. In order to obtain the best possible performance it is desired to remove illumination variations from images. Low rank modeling are often used to model biometrical images as faces, fingers… etc. Low rank + sparse decomposition was recently proposed to capture uneven illumination as sparse errors and was shown to remove illumination variation while capturing the underlying fingers as the low rank component. Here, we propose modeling illumination changes for different fingerprint images as block-low rank, considering that illumination variations are spatially correlated in multiple scales. Initially we adopted an approach to learn a low-rank decomposition for image recognition and then we used Artificial Neural Network (ANN) model to regularize its index of resolvability. The adaptation of a multi-scale low rank modeling, as a matrix summation of block-wise low rank matrices, and increasing scales of block sizes, in addition to the novelty of using ANN in regularizing the index of resolvability, showed enhanced results that under an incoherence condition, the convex program recovers the multi-scale low rank components exactly, which represents the illumination normalization for fingerprint recognition. Experimental results demonstrated the effectiveness of the approach. It showed that the multi-scale low rank decomposition enhanced with ANN regularization the index could provide accurate intuitive decomposition and clearly enhanced results.


Multiscale Decomposition Modeling, Low Rank + sparse Decomposition Modeling, Illumination Variations, Fingerprint Recognition, Artificial Neural Network.