Face recognition is still a challenge in many applications especially in surveillance and security systems. In almost all applications of face recognition, large data sets are used, creating challenges with real time processing and efficiency. In this paper, we propose a hybrid feature extraction method to enhance speed and recognition efficiency. We propose to use features obtained using Histograms of Oriented Gradients (HOG) within a compressive sensing (CS) framework. A HOG feature descriptor algorithm will allow us to extract the face features vector with the advantage of being invariant to changes in appearance or to illumination variations while compressive sensing is used to reduce the density of the resulting face features. This reduction will improve the performance and the computational cost of the system. The classification process is executed using K-Nearest Neighbors algorithm and Probabilistic Neural Network classifier. Our initial experimental results show that the proposed framework is capable of identifying and recognizing faces with different expressions, poses, illumination, and backgrounds in real time.