Combined Local Pattern (CLP): A Novel Descriptor for Face Recognition
Shekhar Karanwal · 2022 4th International Conference on Inventive Research in Computing Applications (ICIRCA) · 2022
LBP is renowned as most powerful texture descriptor. But major issue which LBP possesses is the noisy thresholding function. This sacrifices the discriminativity of the descriptor. To complement that three LBP variants are introduced and these are Mean LBP (MLBP), Median LBP (MnLBP) and Combined Local Pattern (CLP). In MLBP, the mean of the whole patch (3×3) is used for neighborhoods comparison. In MnLBP, the median of whole patch (3×3) is used for neighborhood comparison. The Mean and Median value proves out very effective in contrast to LBP, in which center pixel is used for the neighborhoods comparison. Both MLBP and MnLBP comprehensively conquer the performance of the LBP. To build more productive descriptor features of LBP, MLBP and MnLBP are integrated into one framework. This joined descriptor is called as CLP. The CLP conquer the results of all alone descriptors. Additionally it outrun the performance of the various literature methods. The compression in feature size is accomplished by PCA and classification is assisted from SVMs. Experiments are conducted on ORL and GT face datasets.