Date of Graduation
Statler College of Engineering and Mineral Resources
Lane Department of Computer Science and Electrical Engineering
This research work simulates the human visual cortex by using 2D log polar Gabor wavelet to extract the facial features. Scale and orientation independent convolution of face image with Gabor wavelet gives the features in the form of amplitude and phase. The proposed face recognition algorithm is invariant to frequency, scale, filter orientation, illumination, and contrast. We evaluated the recognition algorithm on four face databases namely FERET, CMU AMP, CMU PIE and Notre Dame Face databases. Experimental results show that using single image for training, phase feature based face recognition performs approximately 5% better than amplitude feature based face recognition.;Another facet of this research involves matching scanned and digital face images. Normalization and transformation algorithms are proposed to resample the scanned and the digital images into one common domain. Validation is performed on a face database of 500 classes containing both the scanned and digital face images.;Finally, a synthetic face database is prepared to evaluate the performance of the proposed face recognition algorithm with disguise. The database includes synthetic face images with single and multiple variations in appearance and feature. Results show that the proposed algorithm outperforms other recognition algorithms.
Singh, Richa, "Unconstrained face recognition for law enforcement applications" (2005). Graduate Theses, Dissertations, and Problem Reports. 1866.