With the technological advances and the evolution of biometric systems, old security methods have became obsolete, increasing attention on researches related to face recognition systems. There are several factors like aging, the use of glasses, hats, makeup and others that make this task nontrivial. This paper proposes a human face recognition system that deals with illumination variation problems in the logarithm Discrete Cosine Transform domain and also improves the classification performance by adaptive feature selection using the the Discrimination Power Analysis (DPA). Experimental results on the Yale B database and CMU PIE database show that our proposed approach improves the performance for the face images with large illumination variations. We also made some experiments with other four databases with different expressions, frontal and profile views and images with cluttered background: the Japanese Female Facial Expression (JAFFE) database, the AT&T database, the UMIST database; and Georgia Tech face database. The experimental results show that the proposed recognition system give superior results compared to recently published literatures.