Radial Orthogonal Mean Local Binary Pattern: An Illumination Invariant Feature Extraction Method

LBP and OC-LBP are the two such descriptors which attains remarkable results for different applications. But under extreme light variations their performances are not adequate. Additionally both possesses the noisy thresholding function. To complement these descriptors the proposed work launches a novel descriptor so-called Radial Orthogonal Mean LBP (ROM-LBP) for Face Recognition. Instead of using raw pixel values for thresholding, the proposed ROM-LBP utilized radials (orthogonal) mean value for thresholding against the center pixel. This radial mean value proves out very dominant than LBP and LBP variants in extreme light variations. Numerous benchmark methods are also outclassed by the proposed ROM-LBP. To compress size FLDA is used and matching is done by SVMs. The entire functionality is tested on two datasets i.e. EYB and YB.