Master of Science (MS)
COMPUTER VISION, FACE RECOGNITION
In computer vision, face recognition is the process of labeling a face as recognized or unrecognized. The process is based on a pipeline that goes through collection, detection, pre-processing, and recognition stages. The focus of this study is on the last stage of the pipeline with the assumption that images have already been collected and pre-processed. Conventional solutions to face recognition use the entire facial image as the input to their algorithms. We present a different approach where the input to the recognition algorithm is the individual segment of the face such as the left eye, the right eye, the nose, and the mouth. Two separate experiments are conducted on the AT&T database of faces . In the first experiment, the entire image is used to run the Eigen-face, the Fisher-face, and the local binary pattern algorithms. For each run, accuracy and error rate of the results are tabulated and analyzed. In the second experiment, extracted facial feature segments are used as the input to the same algorithms. The output from each algorithm is subsequently labeled and placed in the appropriate feature class. Our analysis shows how the granularity of collected data for each segmented class can be leveraged to obtain an improved accuracy rate over the full face approach.
Ghahramani, Farshad, "Face Recognition: An Engineering Approach" (2015). Master's Theses. 4635.