EDRM-LBP: effective directional radial median local binary pattern for face recognition

Recently a novel noise invariant descriptor directional radial median local binary pattern (DRM-LBP) is developed for two applications. In DRM-LBP, median filter is utilised in radial neighbourhood directions for thresholding rather than using raw pixels. Then mean of neighbourhoods median is used for comparison with neighbourhoods for generating DRM-LBP code. After careful evaluation three disadvantages found in DRM-LBP and these are usage of 7 x 7 patch, lack of using centre grey intensity and lack of making fruitful combination for median computation. Proposed work eliminates these demerits and presented novel descriptor effective DRM-LBP (EDRM-LBP) under noisy conditions. In EDRM-LBP, 9 x 9 patch is utilised and then radial neighbourhood pixels are incorporated with centre pixel to generate median value in all directions. In this way five values are utilised for median generation. The obtained median values are then thresholded with mean of those for making EDRM-LBP code. This concept proves very effective in noisy conditions.