Development of algorithms for processing of images of tooth enamel microstructure and fingerprints obtained from distance

Author(s)

Giovani Bressan Fogalli

• Document type: Doctoral Thesis
• Press: Piracicaba, SP.
• Institution: Universidade Estadual de Campinas (UNICAMP). Faculdade de Odontologia de Piracicaba (FOP)
• Defense date: 2023-07-28

Examining board members

Sergio Roberto Peres Line; Liza Lima Ramenzoni; Ricardo José Ferrari; Raquel Fernanda Gerlach; Eduardo Daruge Junior

• Advisor: Sergio Roberto Peres Line

Abstract

Tooth enamel is the hardest tissue in the human body, formed by prism layers in regularly alternating directions. These prisms form the Hunter-Schreger Bands (HSB) pattern when under side illumination, which is composed of light and dark stripes resembling fingerprints. We have shown in previous works that the HSB pattern is highly variable and unique for each tooth. Therefore, it could be useful for personal identification, since the tooth enamel can withstand degradation through time and under aggressive environmental conditions. Although promising, the popularity of this method is hampered by technical limitations in image processing. The aim of the current project was to develop and enable the use of this method through all biometric steps: image acquisition, HSB enhancement, HSB segmentation, filtering and binarization, feature extraction, and matching. Considering the resemblance of the HSB to the human fingerprint pattern, we also adapted and tested the filter used on HSB to touchless fingerprint images. The algorithms of this process are interconnected and therefore were developed and improved together. Results of automated HSB segmentation for two different methods, Anisotropy-based Segmentation (ABS) and Convolutional Neural Network (U-Net), had a Jaccard index of 0.766 and 0.837, respectively. The evaluation of performance of HSB filtering and binarization was performed visually and indirectly by the following steps after segmentation. The feature extraction and matching algorithms were created upon adaptations from fingerprint techniques and revealed as the final result of matching evaluation an Equal Error Rate (EER) of 0.061, when both enlightened sides of each tooth were used together in a sample of 115 extracted teeth. Generally, even though there is room for improvement of applied techniques, the new biometric trait, named toothprint, through these developed algorithms, shows a huge potential for use on a large scale. The adapted filtering step test on finger (AU)

• FAPESP's process: 18/23038-7 - Development of softwares for images optimization of the tooth enamel microstructure in whole teeth and fingerprints obtained from distance
• Grantee: Giovani Bressan Fogalli
• Support Opportunities: Scholarships in Brazil - Doctorate