Gaussian mixture models for forensic applications

Recently, significant progress has been made in designing fingerprint identification systems for missing fingerprint information. However, a dependable reconstruction of fingerprint images still remains challenging due to the complexity and the ill-posed nature of the problem. One of the most important areas in biometrics is generating synthetic fingerprint databases and matching the partial fingerprints. The goal of this thesis is to develop a system based on Gaussian mixture models for forensic applications. In this thesis, we present a duplication algorithm for an image which can be used for generating synthetic fingerprint databases and matching the partial fingerprints. Computer simulation shows that the duplication process is more "universal" and also can be used for roof image duplication. The performance of the presented algorithm was evaluated by Structural Image Similarity Measure, Entropy and Mean Square Error. In addition, we reconstruct the missing data from both binary and gray-level fingerprint images and present a new similarity score to evaluate the performance of the reconstructed binary image. Moreover, we present texture transform algorithm by using multiple and single roofs images. The performance evaluation was done using visual statistical methods. The developed algorithms offers a potential for further research and application in other surface inspection processes for generation of synthetic databases which can be used in forensic applications.