Face analysis is an emerging area in biometric research. This paper proposed the robust local descriptor using deep learning for face analysis. Recently the researchers used the adaptive filter approach to enhance the image instead of using the static filter approach. Deep learning helps to create an adaptive filter and it achieved significant results on face recognition. This motivated us to use deep learning concepts to recognize the face. In this paper, dimensionality-reduced local directional pattern descriptor is used to enhance the face image. PCANet model is applied over the resultant encoded image to discriminate the features of face the image that is used for identification and classification of the face image. The experiment was carried out with the standard benchmark databases such as FERET, Extended Yale B, LFW, ORL and AR databases. This method achieved better recognition rate compared with state-of-the-art methods. © 2017, Springer-Verlag London Ltd., part of Springer Nature.