PERFORMANCE OF FACE RECOGNITION WITH PREPROCESSING TECHNIQUES ON ROBUST REGRESSION METHOD

Abstract

The Robust Regression method has been used successfully in face recognition problems.
Based on empirical experiments on some standard face image databases, the method shows very high
accuracy. The method used the histogram equalization technique to normalize illumination such that the
effect of illumination factors is reduced substantially on the image. In this research, some contrast adjustment
techniques are used in the pre-processing stage to determine how far those techniques affect the face
recognition performance. There are three contrast adjustment techniques used, i.e. Histogram Equalization
(Histeq function), Contrast-limited Adaptive Histogram Equalization / CLAHE (Adapthisteq function) and
Imadjust function. In addition, it is also used the no-pre-processing technique (not using pre-processing
techniques). The experiments were performed on three standard face image databases, i.e. CMU-PIE Face
Database, Extended Yale Face Database B, and AR Face Database. The experimental results show that the
use of Adapthisteq function in the pre-processing stage of the Robust Regression method produces the
highest average accuracy of 97.69%. This result is better than the accuracy of Histeq, Imadjust, or no-preprocessing technique, which are 94.53%, 90.59%, and 93.43% respectively.