A novel approach for matching of dental radiograph image using Zernike moment

Recent growth in the number of studies for automating the process of postmortem (PM) identification of deceased individuals based on teeth characteristics is remarkable. The enhanced scalability of this biometric is more remarkable when we pay more attention to the number of victims involved in mass natural disasters (e.g., airplane crashes, Asian Tsunami), and urgent need to a reliable method for automating the identification process. Consider a dental radiograph image of a postmortem, this study shows us how a search will be done in a database of antemortem (AM) radiographs in order to find the best matches with attention to the most significant features. In this study we present a human identification system, which uses a new method named Zernike moment (ZM) as a feature extractor and the Euclidian distance for classification in matching stage. In previous published works, the contour of teeth has been used as a feature for matching. In these methods there were failures in some results due to dental tooth extraction which sometimes causes distortion and inclination of teeth towards each other. So it leads to improper identification results. But in this study, we extracted features of teeth with Zernike moment which is invariant in rotation and scaling. In this work after the segmentation step and then the numbering of teeth according to the universal numbering system, we first extract high-level features to reduce search space. The next step is extracting low-level features for each tooth to find the closest matched tooth in the database to the query tooth using the Euclidian distance. Another new idea in this paper is to use the majority votes of dental records. In the other word, for each record of dental radiograph image we have a few teeth to consider which tooth has the most votes as a closest match. This will enable us to recognize each query record belongs to which AM record. Experimental results obtained for this investigation on a database of 560 AM and PM teeth shows encouraging result in detecting true record for query image.