<p>Makeup Invariant Face Recognition is a global problem from the perspective of security and recognition. It has various applications like criminal identification, driving license or passport verification, etc. Makeup variation significantly alter the appearance of the face by affecting the texture, shape, contrast of the mouth and eye regions. This causes the inability of traditional face recognition methods to consistently identify individuals. The deep learning methods have improved performance; however, these are computationally intensive and require large annotated datasets. Therefore, to address the above problems, the objective of the paper is to propose a novel methodology DGLDD-FR which handles makeup variation using a local derivative descriptor and four directional gradient images. Initially, the image is preprocessed, followed by convolving four directional masks over the grayscale image to generate the gradient images. Subsequently, the local derivative descriptor is applied to the regions of the gradient images, allowing for the extraction of discriminative features. Finally, a Support Vector Machine is employed to recognize the face images of different individuals. The effectiveness and robustness of the proposed methodology have been evaluated on standard datasets YMU and VMU, and achieved accuracy 91.25% and 91.50% respectively. The proposed methodology outperforms existing discriminative methods (weber faces + LGBP by 0.61%, Gradient faces + PCA by 6.75%) and deep learning approaches (FGGNet by 1.21%) on YMU dataset and outperforms discriminative methods (LGBP by 9.30%, LBP by 13.20%, HOG by 17.90%) and deep learning methods (FGGNet by 3.02%, LSTM by 5.50%) on VMU dataset.</p>