Research On Methods Of Similarity Measurement In Computer-aided Photographic Superimposition

Photographic superimposition is the primary means of comparing unknown skull and photos of missing people to identify the true identity of the skull. It could provide clues for criminal detection and investigation of disastrous accidents, with high research value and practical significance. Computer-aided photographic superimposition’s key step-similarity measurement of skull-photo overlay is studied and discussed in this paper, a new method of similarity measurement is proposed, which solves the problems such as single index and too much manual intervention. The speed and accuracy of identification are improved. The main researches on this thesis include:(1) A set of new metrics of similarity measurement on superimposed photos are proposed. Refer to the forensic anthropology theory, the researches of multiple superimposition researchers and processing requirements of computer,41 indicators are proposed in this paper, including indicators of feature points, indicators of index, indicators of soft tissue thickness and contour curves. Besides, the portrait shooting angle’s influences on these indicators are studied to improve the accuracy of identification. The distribution of these indicators is adequate and reasonable, laying the foundation for the establishment of comprehensive and synthesized similarity measuring method.(2) A set of measuring methods on above indexes are proposed. Based on the principle of regression and forecasting, regression models of upper and lower bounds on indicators of feature points, index and soft tissue thickness are established to measure these indicators; the method of curve’s similarity measurement based on Fourier descriptors is adopted to measure the similarity of mandibular curves. Finally, the number of items that meet the requirements of first three indicators and the similarity of curves are 4 inputs of new measuring method. Experimental results show that the measuring method is fast and efficient with high accuracy.(3) A new method of similarity measurement on skull-photo overlay based on BP neural network is proposed. The improved BP neural network(momentum-adaptive learning rate algorithm) is adopted in this method to train the network and measure rules based on four categories of evaluation indicators are established. Use the rule to determine whether the skull and photo belong to the same person. Experiments show that the new measuring method which integrates different indicators is reasonable and objective, with high accuracy and high speed of training and judgment.(4) The system of computer-aided photographic superimposition is designed and realized. The system includes many features, for example:file read, interactive operation, feature points extraction, contour line extraction, skull-photo overlay and similarity measurement etc. The system is applied to real databases of missing people, improving the identification efficiency of unknown remains.