Computer-aided processing techniques for usage in real-time image evaluation

The capture -and conviction of criminals is often hampered by the present cumbersome method of suspect identification. A witness to a crime may have to view thousands of photographs in order to find the one that corresponds to the suspect. This process can be speeded up by utilizing a computer-aided system. This thesis provides a computer program that takes as an input a digitized version of the suspect's photograph or sketch. This information is processed by the computer and provides as output, the location of the major features(eyes, nose, mouth, etc.) and also a facial outline. Using this information an algorithm can be developed that can be used to process the photographs contained in a police department's mug file by producing an array containing information that is unique to each photograph. Then when a crime is committed, the digitized version of the suspect's sketch can be given to the computer to construct this array. The array can then be compared to those on file and those most similar can be chosen. Then the witness would only have to view these few photographs. Before the development of this automatic feature locating algorithm facial measurements were taken by hand at a rate of 5 minutes per case. Using this algorithm, the major facial features were accurately located within 10 seconds in over 88% of the test cases. The facial outline that was produced by the algorithm was acceptable in 66% of the cases. When implemented in a user interactive system, the algorithm presented in this thesis should significantly speed up the process of suspect identification.