Making face recognition more reliable under uncontrolled lighting conditions is one of the most important challenges for practical face recognition systems. Data preprocessing thus becomes an important and emerging topic in many data-driven applications such as image processing and bioinformatics. Dimensionality reduction provides an efficient way for data abstraction and representation as well as feature extraction. It aims to detect intrinsic structures of data and to extract a reduced number of variables (dimensions) that capture and retain the main features of the high-dimensional data. For instance, images contain a large number of pixel values and are presented as high-dimensional arrays. The computationally efficient combination of the most successful local appearance descriptors, like Local Binary Pattern (LBP) with its extension Local Ternary Patterns (LTP) for facial appearance and Gabor filter to encode facial shape over a range of coarser scales are implemented. Here, a data mining approach for dimensionality reduction provides an efficient way for data abstraction and representation as well as feature extraction. It aims to detect intrinsic structures of data and to extract a reduced number of variables (dimensions) that capture and retain the main features of the highdimensional data. The resulting method provides state-of-the-art performance on different data sets that are widely used for testing recognition under difficult illumination conditions: Ex-tended Yale-B, CAS-PEAL-R1. Further experiments show that our preprocessing method outperforms several existing preprocessors for a range of feature sets, data sets and lighting conditions by comparing with previously published methods, achieving a face verification rate of 89.1% at 0.2% false accept rate.