This thesis presents an approach for personal identification/authentication with the use of Hand Geometry. Some related work has been done in this area, but mostly pegs are required when capturing hand images and only the upper half of the hand area is used for encoding a feature vector. We present a new feature extraction method that utilizes the whole hand and can handle a variation of hand placements. New features such as hand area and hand contour length are found to be effective in identifying person. No pegs are needed to fix the hand position and no contact to the capture device is required. The captured image will be processed to form binary (black and white) and line images. A Feature vector with 29 features will be extracted from these two images. Further feature selection pro...[ Read more ]

This thesis presents an approach for personal identification/authentication with the use of Hand Geometry. Some related work has been done in this area, but mostly pegs are required when capturing hand images and only the upper half of the hand area is used for encoding a feature vector. We present a new feature extraction method that utilizes the whole hand and can handle a variation of hand placements. New features such as hand area and hand contour length are found to be effective in identifying person. No pegs are needed to fix the hand position and no contact to the capture device is required. The captured image will be processed to form binary (black and white) and line images. A Feature vector with 29 features will be extracted from these two images. Further feature selection processes reduce the size of a feature vector to 17. Noisy samples can be eliminated using the outlier removal process. We present promising results, which show up to a 98.7% successful classification rate and a 2.8% Equal Error Rate on a hand image database comprising one hundred individuals.