Efficient face recognition using frequency distribution curve matching

To develop an accurate and efficient face recognition system, a technique is proposed that preserves holistic as well as local facial details. It employs the cumulative frequency distribution curve (FDC) of the grey levels and their standard variance. Based on the FDC, a reduced space profile is produced, which is composed of three distinct segments. Each segment has its own associated error value defined in the range (0, 1). A decision is made by evaluating error values between the training and testing data sets independently for each segment in the FDC. A face is recognised accurately if there is conformance in all of the three segments; that is, the relevant error conditions are met simultaneously in all the segments. With one-shot training, the proposed technique is not only faster but also provides 14, 1.9 and 1.7% improvement in accuracy for the Yale, Olivetti Research Laboratory and pose, illumination and expression databases compared with other widely used face recognition techniques.