以牙齒與補牙形狀為特徵之自動化牙齒身份辨識系統

Abstract

自動化牙齒身份辨識系統(Automated Dental Identification System, ADIS)在法證界中是一個相當重要的身份辨識技術，主要目的是用來確認身體組織遭受到嚴重破壞的罹難者或是失蹤人口的身份。在本篇論文中，我們提出了一套基於牙齒X光咬翼片之自動化牙齒身份辨識系統，其中包含:牙齒X光片影像增強(enhancement)、影像切割(segmentation)、牙齒分類(classification)、牙齒編號(numbering)、以及牙齒辨識(identification)等功能。 在影像增強階段，我們結合同態濾波(homomorphic filtering)、自適性對比延展(adaptive contrast stretching)以及自適性形態學轉換(adaptive morphological transformation)等方法來同時增強牙齒X光片的對比度以及改善曝光不一致的問題。在影像切割階段，我們提出結合疊代門檻值法(iterative thresholding)以及疊代積分投影(iterative integral projection)法來進行牙齒與牙髓的切割與分隔。並且利用一個兩階段式的牙齒治療(dental work)切割法來進行補牙(filling)之萃取。在牙齒分類階段，我們提出一個牙齒歪斜糾正(skew-adjusted)的方法來有效取得正確的牙齒與牙髓相對長寬比(length/width ratio)，以及搭配相對的牙齒大小作為特徵值進行牙齒分類。在牙齒編號階段，我們結合一個缺牙偵測(missing teeth detection)演算法和一個序列比對(sequence alignment)方式來給予每顆牙齒一個特定的編號。最後，我們利用牙齒以及補牙之形狀作為比對之依據，進行牙齒身份辨識。為了減少不適用(unreliable contour)和不完整(incomplete contour)之邊界所造成的對位錯誤(alignment error)，我們提出一個有效邊界之距離權重(effective contour alignment based on weighted Hausdorff distance)對位法來提升比對之準確率。同時，為了彌補在空間域中對位不完全(imperfect alignment)的問題，我們利用具有對位不變性(alignment-invariant)的頻率域特徵作為互補。實驗結果證實我們所提出之方法確實能夠有效的提升分類、編號、以及辨識之準確率。

Automated dental identification system (ADIS) is an important human identification system designed for law enforcement to identify victims or missing people who are under severe decaying of soft tissues of the body. In this dissertation, we establish an ADIS to effectively enhance, segment, classify, number, and identify the teeth in dental bitewing radiographs. An image enhancement method that combines enhanced homomorphic filtering, homogeneity-based contrast stretching, and adaptive morphological transformation is proposed to improve both contrast and illumination evenness of the radiographs simultaneously. Iterative thresholding and iterative integral projection are adapted to isolate teeth to regions of interest (ROIs) followed by contour extraction of the tooth and the pulp from each ROI. A dental works (DW) segmentation method based on the two-stage algorithm is adopted to extract the contours of DWs. A binary linear support vector machine using the skew-adjusted relative length/width ratios of both teeth and pulps, and crown size as features is proposed to classify each tooth to molar or premolar. A numbering scheme that combines a missing teeth detection algorithm and a simplified version of sequence alignment commonly used in bioinformatics is presented to assign each tooth a proper number. Finally, an enhanced dental identification method based on both the contours of teeth and dental works is presented. For reducing the alignment error caused from unreliable contours, a point-reliability measuring method is proposed when calculating Hausdorff distance between the contours. For reducing the alignment error caused from incomplete tooth contours, an outlier detection method is proposed to prune the outlier from each contour and realign the pruned contours. And for compensating the error when matching with the spatial feature of DWs due to imperfect alignment of the teeth in which they reside, an additional alignment-invariant frequency feature of DWs is used. Experimental results show that our proposed system can achieve (1) 95.1% and 98% for classification and numbering, respectively, in terms of number of teeth tested, and correctly classifies and numbers the teeth in four image that were reported either misclassified or erroneously numbered, respectively; (2) 94.3% and 100% image retrieval accuracy of the top-1 and -5 retrievals, respectively, when matching with weighted HD of the pruned contour of a single tooth, which is superior to a method that matched with the traditional HD of two or three adjacent teeth; (3) 100% accuracy of top-2 (top 6%) image retrievals when matching with both contours of teeth and dental works, which again is superior to a method that matched with the size of non-registered tooth contours and non-overlapped dental works of a pair of teeth.