In this paper, we propose a hybrid statistical feature extractor, Local Haar Mean Binary Pattern (LHMBP). It extracts level-1 haar approximation coefficients and computes Local Mean Binary Pattern (LMBP) of it. LMBP code of pixel is obtained by weighting the thresholded neighbor value of 3×3 patch on its mean. LHMBP produces highly discriminative code compared to other state of the art methods. To localize appearance features, approximation subband is divided into M×N regions. LHMBP feature descriptor is derived by concatenating LMBP distribution of each region. We also propose a novel template matching strategy called Histogram Normalized Absolute Difference (HNAD) for histogram based feature comparison. Experiments prove the superiority of HNAD over well-known template matching techniques such as L2 norm and Chi-Square. We also investigated LHMBP for expression recognition in low resolution. The performance of the proposed approach is tested on well-known CK, JAFFE, and SFEW facial expression datasets in diverse situations.