Dictionary based learning has emerged as a powerful approach to a large class of machine learning problems, especially face recognition. The development of face recognition methods for unconstrained environments is still a challenging problem. In this article the authors present a dictionary based approach that considers compact face features to define a cluster centroid using k-means clustering in conjunction with a sparse representation classifier. The varying environmental aspects of human face recognition, namely, illumination and facial expression, have been dealt with for images captured under controlled and uncontrolled settings. Face normalization using the gradient face method is employed to handle variations in illumination conditions. Facial expression is handled by the use of compact face features, generated using the popular rotation invariant uniform local binary pattern and the histogram of gradients. The efficiency of the proposed method is demonstrated using three large benchmark databases with vast variations, extended Yale B, CMU-PIE and IMFDB. It is encouraging to note that the proposed method has superior performance to popular face recognition algorithms.