An Effective Dental Work Extraction and Matching Method for

Abstract

This paper presents a method of dental work (DW) extraction and matching for bitewing radiographs. The method includes an algorithm that can correctly extract DW’s from radiograph images and two matching metrics (dFD of frequency domain and dXOR of spatial domain) that can effectively match a postmortem DW to the genuine antemortem DW in the database. The experimental results show that our method can extract DW’s more correctly shape- and number-wise than methods that use global thresholding. The results also show that, for a database containing 52 different shaped DW’s, the accuracy based on dFD is 95% when using top 6% retrievals and achieves to 100% when using top 17% retrievals; whereas the accuracy based on dXOR is 86 and 95 %, respectively, when using the same percentage retrievals. For a database containing 10% of hoax DW’s (i.e., DW’s have the same shape but different orientation/position as some genuine DW’s), the accuracy based on both metrics maintains about the same except the top-1 match based on dFD. The fact that dXOR was not affected by hoax DW’s indicates that dFD and dXOR can compensate each other when matching the postmortem DW with hoax DW’s. Comparing with the 95% and 100% hit rates using top 8% and 47% retrievals from a much bigger database reported in literature, the metric dFD not only is more efficient but gives a better result.1