Research On Characteristic Parts Selection And Recognition Of 2-D Skull Image

With the development of information technology and network technology, people pay much more attention to information security. Personal identification is one efficient way to protect information system. It has been used in many fields, such as financial department, national security, and judicial systems etc. Biometrics is a technology, which uses people's physiological characteristics to identify a person. Recognition system extracts feature from people's physiological characteristics and saves the feature into database. When a person uses the system, his feature will be compared with the features, which have been saved in the database, to testify who he is.The research in this dissertation is based on reconstructed CT images. Basic knowledge of forensic anthropology and some feature exaction methods have been studied. The main work includes:1. The contour of skull, orbit and its contour, the curve of submaxilla have been selected as the important characteristics of the skull.2. Method of image feature extraction based on SVD has been used to extract the character of whole skull images. The feature of orbit has been tested using image registration algorithm. Mutual-information maximization has been used to separate orbit features. Experiments show that the gray information could be used in skull recognition while images are normative.3. The curvature of submaxilla's curve has been studied in this dissertation. Correlation of submaxilla's curve has been test, and experiments show that curves of submaxilla from the same person are correlated, while the curves of submaxilla from different persons are uncorrelated.4. The feature of skull contour has been extracted by using polar -radius-invariant-moment, Fourier descriptors and wavelet descriptors separately. The characteristic parameter for orbit contour has been constructed. The recognition results show that polar-radius-invariant-moment and wavelet descriptors are effective in skull and orbit contour feature extraction.In this dissertation, character and algorithm has been studied. Plenty of tests have been done to validate the efficiency of characteristic parts, and the result show that the character and algorithm we choose are favorable to the recognition.