M. Hassan, I. Osman, and M. Yahia (Sudan)
Face recognition, Discrete Sine Transform.
This Paper proposes a face recognition model that depends on combination between Discrete Sine Transform (DST) and Alternative Local Linear Regression approach (ALLR), we use DST for facial feature extraction and ALLR applied for generate virtual frontal faces from non-frontal images, the images selected from face database of Olivetti Research Laboratory (ORL). The paper presents the optimal mapping function for finding the DST coefficients that increase the recognition rate, also the paper show that the non-absolute linear functions are better than absolute non-linear functions when applied with DST. Many experiments were conducted, the ALLR obtained 16.85% improvement in recognition rates over DST average of many different experiments over the DST, the Back propagation neural network (BPNN) obtained 2.5% more improvement over ALLR in recognition rate, finally the combined algorithm (DST+ALLR+BPNN) resulted in 100% recognition rates.
Important Links:
Go Back
