![]() | Only 14 pages are availabe for public view |
Abstract This Thesis presents two algorithms for face recognition. The first algorithm is based on the fast Fourier transform (FFT) technique and the second utilizes the discrete cosine transform (DCT). These transforms were applied to the low-frequency components of the wavelet decomposed images either to the third or the fourth level. They are simple, do not suffer from computational complexity, and handle different variations in face images such as different features (smiling - crying – sleeping- … etc.), illuminations, small occlusion (wearing glasses), poses, scales and rotations (face rotated with different angles). Three image feature matrices were utilized depending upon the pose of the input images used. Classification was performed using the minimum distance classifier. Experimental results indicate that the FFT algorithm works well with lighting variations. The DCT algorithm, on the other hand, is insensitive to face position as was detected when using rotated face images to about 15 and to face variations. The performance of the suggested algorithms was compared with those obtained by template method [8] and eigenface method [18] and was found to be significantly more satisfactory. They have the ability to overcome differences in face images whether resulting from different features, illumination or small occlusion. |