Fractional Discrete Cosine transformation based reduced set of coefficients for face recognition

In this paper an attempt is made to explore the effect of using reduced set of Discrete Fractional Cosine Transformation based features on the face recognition accuracy. Input image feature set is transformed from spatial domain to spatial frequency domain using FRDCT. The large number of coefficients of fractional order spectrum of the face images obtained by the application of 2D FRDCT is scaled down by classical data dimensionality reduction technique LDA approach. Reduced feature set is then classified by the use of nearest neighbor classifier. The effectiveness of the proposed approach is demonstrated through the simulation on the benchmark database (AT&T). Experimental results also show that unlike DCT, which preserves strong information packing capability, FRDCT also preserves this capability with varying rotation orders.