Abstract—Forensic dentistry generally addresses the problem of identifying individuals based on the some specific characteristics of teeth or bite mark impressions. Bite mark identification process generally involves human interaction and has human bias. It would be beneficial to have a system that reduces human bias and has high accuracy matching performance. This paper describes a preliminary study to verify the effectiveness of applying the neural network approach in bite mark identification. By selecting some specific features of the bite marks for the model, trained networks give reasonable result for the matching accuracy in this initial study. Index Terms—Bite mark identification, forensic dentistry, image analysis, neural network application I. INTRODUCTION ite mark is a type of evidence which may be found after a crime. However, this type of evidence is controversial and needs further study. Many methods for bite mark identification have been proposed ranging from manual approach, semi-automatic approach, toward fully automatic approach in recent years [2-5]. Several algorithms for comparison process have been applied to improve the accuracy of the identification. The Artificial Neural Network (ANN) is an adaptive system which changes its structures based on external and internal information that flows through the network during the learning phase, and is suitable for analyzing complicated relationship [1]. A review of the literature identified three studies in which ANN were applied to caries detection in teeth and human craniofacial growth analysis and classification [6-8]. This preliminary study was aimed to verify the success of applying ANN in bite mark analysis by testing 34 specific features of bite marks which were registered on pink wax to have a clear view of the bite mark. Manuscript received December 28, 2010. This work was supported by Faculty of Dentistry, Chiang Mai University, Chiang Mai, Thailand. P.M. Mahasantipiya is with Department of Oral Biology and Diagnostic Science, Faculty of Dentistry, Chiang Mai University, Chiang Mai, 50200, Thailand. U. Yeesarapat, T. Suriyadet, J. Sricharoen, and A. Dumrongwanich are dental students at Faculty of Dentistry, Chiang Mai University, Chiang Mai, 50200, Thailand. T. Thaiupathump is with the Computer Engineering Department, Faculty of Engineering, Chiang Mai University, Chiang Mai, 50200, Thailand. (corresponding author, phone: +66 86 6543212; e-mail: trasapong@eng.cmu.ac.th) II. MATERIALS AND METHODS Data Collection Bite mark samples were acquired from 50 Thais with an age range of 18-25 years. The assumptions in this study for each sample are no missing any of lower and upper anterior teeth and no fixed orthodontic appliances. Each sample was asked to bite on standard pink dental wax with five different biting positions which chosen from the following list; 1) faced front standing, 2) face front sitting, 3) chin down standing, 4) chin down sitting, 5) chin up standing, 6) chin up sitting, 7) supine position, 8) prone position, 9) left bed side position, 10) right bed side position. Bite marks of all upper and lower anterior teeth were covered. Next, Photographs of these bite marks were captured with a digital camera Panasonic LX3 (Panasonic Corporation, Osaka, Japan). The resolution is 11.3 million pixels and the distance between the objects and lens was 25cm. These photos were stored as JPEG files as bite mark samples. Thus, there were a total of 250 bite mark samples and only bite marks from lower teeth were used for ANN training in this study. Pre-processing techniques Each bite mark sample was transformed from bit depth of 32 to 8-bit gray scale (0-255), where 0 is the darkest and 255 is the brightest. Later, each image was adjusted its contrast by using Contrast Limited Adaptive Histogram Equalization. Region of interest covering only all anterior teeth was cropped. Then the median filter was applied to enhance the border of each tooth bite mark, followed by converting the image to black and white format, as illustrated in Fig.1-3. Fig.1. Initial bite mark image Feature Selection Accuracy level of the identification process depends heavily on the appropriateness of the selected features. In this preliminary study, some criterions need to be Bite Mark Identification Using Neural Networks: A Preliminary Study P.M. Mahasantipiya, U. Yeesarapat, T. Suriyadet, J. Sricharoen, A. Dumrongwanich, and T.Thaiupathump B Proceedings of the International MultiConference of Engineers and Computer Scientists 2011 Vol I, IMECS 2011, March 16 - 18, 2011, Hong Kong ISBN: 978-988-18210-3-4 ISSN: 2078-0958 (Print); ISSN: 2078-0966 (Online) IMECS 2011