| Postmortem classification is the vital task in forensic field. Most of methods based on living matters such as lip print, fingerprint, iris, and so on, however, can hardly make it or even absolutely play out in case of some catastrophe such as malignant case, explode, air crash, tsunami, fire or traffic accident.Dental feature will play the unique role in these intractable cases mentioned above. On one hand, teeth are the hardest organization in human body, with high resistance to elevated temperature and erosion. On the other hand, the probability that different individual carries the same dental feature is just a2,500millionth, which enables the dental feature to be another "ID series", as the dentists confirmed. Such stability and diversity will make the classification method based on dental feature the most potential and valuable tool in individual identification. Actually, many national justice bureaus have confirmed the importance of forensic dentistry in forensic science.Dental feature can be recorded as antemortem information by dental radiograph in order to be compared with the postmortem ones in the future. So it’s beyond the dispute that every method of individual identification based on biological characteristic, including dental feature, is essentially a branch of content-based image retrieval, and the common way of these tasks is the combination of essential theory of image retrieval with the correspond kind of biological characteristic. Therefore, effective means of feature extraction and registration become the key point in this article.Each main component of the dental identification system designed in this article is listed as following:1) Establishment of Medical Teeth Radiograph Database. The designed dental identification system, as a derivation of content-based image retrieval, must be equipped with an image database, and the database is composed of more than100dental radiographs.2) Image Preprocessing. Medical radiograph processing is usually exacerbated because of its poor quality such as noise and uneven lightning. The linear gray-level enhancement have been performed in order to improve image quality and suppress all kinds of noise.3) Feature Extraction. Feature extraction is the vital step in data analysis. Modern image feature is composed of gray-level feature, edge feature, and transformational argument feature and so on. Given the stability of human teeth and optical nature of dental-work area, key-point model and contour model is selected as image feature and illustrated in this article.4) Feature Matching. The main task of this step is to select a measurement of feature matching. Several measurements are selected as the indicators of dental feature matching, and the values of these measurements will get to extrema when the two features can be correctly matched.Image feature extraction and feature matching, as the two emphases among aforementioned components, will be detailed. And in order to implement the components, some algorithms, including topology point, feature contour, model estimation based on RANSAC (Random Sample Consensus) and their application to dental image processing and registration, is researched in this article.1) Combination of SURF (Speeded-Up Robust Feature) key-points extraction and RANSAC model estimation is performed to extract the SURF key-points from input image and reference image as image-matching feature, and to estimate the graphic transformational model according to the result of image matching, thus enhancing the accuracy of the whole system.2) A new RANSAC algorithm base on the correlation parameter is proposed aimed at the flaw of conventional RANSAC algorithm, thus reducing the effect of noise to some extent, and enhancing the reliability of the whole system.3) Fourier descriptor is applied to express the feature of dental-work contours and used to match, thus improving the matching performance. |
PDF Full Text Request