Analysis Of Dental Image Processing For Human Identification

DOI : 10.17577/IJERTV1IS10322

Download Full-Text PDF Cite this Publication

Text Only Version

Analysis Of Dental Image Processing For Human Identification

Anita Patel, Pritesh Patel, Assi.Prof. Ashtha Baxi

Parul Institute Of Engineering Technology, Baroda, Gujarat, India


Dental X-ray has played an important role in human identification. Particularly, in cases of severe accidents, plane crash, earthquake, etc. Wherein other identification clues like fingerprint, iris, etc. are not available for identification of missing or unidentified persons and moreover dental features remain more or less invariant over time. The purpose of dental image processing is to match the post-mortem (PM) radiograph with the ante-mortem (AM) radiograph based on some characteristic or feature of the radiograph. The dental images acquired may be of poor quality and contrast. Hence it is important to first enhance the quality of image and thereafter various segmentation algorithms are applied to the enhanced dental image. Various features of individual tooth are being extracted and identification is performed based on matching of these feature vectors of

PM images with those of AM images. This paper gives an overview of various enhancement, segmentation and feature extraction, matching technique for human identification using dental radiographs.

Keywords-Antemortem (AM) radiographs, Postmortem (PM) radiographs, segmentation, Human identification, Dental images.

  1. Introduction

    Automated dental identification system (ADIS) was developed for extraction of the distinctive features of tooth images [9]. They were then compared automatically to each other and searching for the best candidates in the database to identify the unknown person. The main process of this system contains of image classification, teeth segmentation, feature extraction and pattern matching. Fig. 1 shows the process of this scheme.

    The first step in human identification is dental image classification which is based on the way dental features are captured. They are classified as bitewing, periapical and panoramic dental images [10] as shown in figures (2) to (5). Bitewing images include the features of both jaws signifying bite. While periapical images include only a single jaw either upper jaw called upper periapical image or lower jaw called lower periapical image. Panoramic images include features of both jaws including sinuses, nasal area, etc.

    The main process of ADIS [9]

    However, for most dental processing bitewing images are used [10]. The dental radiograph can be divided into background area in lowest intensity, Bone areas in average intensity and teeth areas in highest intensity. In some cases, the intensity of bone area and teeth area are nearly same. So, they should be separated for fruitful feature extraction. The next step is radiograph segmentation which includes separating upper and lower jaw and thereafter separating each individual tooth. Feature Extraction process follows

    tooth segmentation wherein some specific features are defined (like curve, shape, texture, etc.) which is further used for matching PM images with AM images.

    The next step is matching of AM and PM radiograph. A matching distance is found for each pair of PM-AM images based on matching technique. Depending on the matching distance, the images from the database are ranked. The minimum matching distance image is found as the best match of given PM image. The accuracy rate of algorithm is found from the rank obtained by the most genuine AM image. Ex more the number of PM query images having lower rank of the genuine image, higher is the accuracy rate. For comparing various techniques we have defined percentage of genuine images ranked as first as follows:

    % Perf=A * 100 /P

    Where % Perf=performance index; A=no. of geniune AM images retrieved having Rank 1 &P=no. of PM query images on which technique was implemented.

    Fig. 2. Bitewing dental image [10]

    Fig. 3. Upper periapical dental image [10]

    Fig. 4. Lower periapical dental image [10]

    Fig. 5. Panoramic dental image [10]


    Thresholding based methods can easily be used for segmentation of teeth but they usually fail to classify between teeth and bone areas as their intensities are more or less similar in cases of uneven exposure. Anil K. Jain have suggested Tooth Contour Extraction for Matching Dental Radiographs[1].

    Automated Dental Identification System (ADIS) requires not only identification of the subject but also maintaining the system such as updating reference records, updating techniques and substandard performance [2]. In order to overcome the difficulty arising due to poor quality of image, in [2004], M. Mottaleb using iterative and adaptive thresholding. Thereafter horizontal and vertical integral projection is used for separating the jaws as well as individual tooth. The case in which jaws are not aligned along a horizontal line the image was rotated in a small range of [- 20,20]degrees and the angle which produces minimum horizontal projection is found. A set of salient points from object contour is selected & a signature vector that captures information of each salient point is generated. Each element in Signature vector is the distance between the salient point and point on the contour. Matching distance is then found from the signature vector & ranking based on minimum matching distance is performed. Best matching AM tooth correspond to minimum matching distance. This technique was not successful in matching images due to poor quality of images;

    shape of teeth could have changed with time as PM images were taken after a long time AM images were captured and handling view variance in both AM & PM images. The %Perf achieved by this method is 72.41%.

    In [2005], H.Chen and A.K.Jain at aligning the partial contour in case of occluded image is addressed and contours of tooth as well as shapes of dental work are used for identification. Upper and lower jaw as well as individual tooth is separated using horizontal and vertical integral projection as in [2]. Difference between contour of teeth and difference between contours of dental work are combined via likelihood estimates for better similarity results. Matching is done by computing the matching distance between one PM and all AM images & then image to subject distances are averaged over all images to obtain matching distance. From the distance between PM image & all subjects in AM database, ranking generates a list of candidates. However, this technique was not capable to produce desired results in cases of poor image quality; subjects with missing teeth and it moreover it required a larger database for evaluating the algorithm.

    In [2006], O.Nomir and M. Mottaleb at suggested a hierarchical chamfer matching methods for contour matching. At using segmentation method starts by applying iterative thresholding followed by adaptive thresholding to segment the teeth from both the background and the bone areas [4]. After thresholding, horizontal integral projection followed by vertical integral projection is applied to separate

    each individual tooth. Here used by edge matching algorithm that uses the Hierarchical Chamfer Matching. It is a technique for finding the best match for a given image by minimizing a predefined matching criterion in terms of the distance between the contour points of two images. To increase the accuracy of matching, reduce the search space and computational load. The advantage of this technique is that the matching is applied using multiresolution algorithm. The %Perf achieved by this method is 80%.

    In [2007], O. Nomir and M. Mottaleb by exploring the appearance and shape- based features [5]. Here, the image is first enhanced by binary image masking & thereafter adaptive thresholding is applied. After adaptive thresholding, horizontal integral projection followed by vertical integral projection is performed to separate individual tooth. The contour is extracted by Fourier descriptors are powerful for two-dimensional shape description. It used by force field energy function. Matching is done using both L1 norm (absolute distance) and L2 norm (Euclidean distance).In both the cases, majority voting is used to obtain the best matching image, because in some cases, there is a large distance between one or two corresponding PM and AM teeth for the same person due to poor image quality, which increases the total matching distance resulting in incorrectly extracted contours of same teeth. The %Perf achieved by this method is 86%.

    In [2008], S. Kiattisin proposed algorithm for 2 features of teeth for code matching namely labial view (having one

    root) and mesial view pattern (having two roots). Brightness Adjustment; Binary image Conversion were used for image enhancement[6]. Chain code method was used for decoding a direction code from binary images based on special features of teeth. Special features of teeth were extracted as a feature. Matching is done by comparing the decoding code with the statistical code. However, the resulting chain of codes tends to be quite long and moreover, any small disturbances along the boundary due to noise or imperfect segmentation causes change in code that may not be related to the shape of the boundary. The %Perf achieved by this method is : code match=90% for same code & 50% for different code.

    In [2009], P. L. Lin and Y. H. Lin propose a dental classification system to effectively classify molar teeth from premolar teeth in dental bitewing radiographs [7]. In system include a novel image enhancement method that combines homomorphic filtering technique to reduce the uneven exposure problem, both adaptive contrast stretching and adaptive morphological transformations based on homogeneity to accentuate the texture differences between teeth and gums and between teeth and pulps. The shapes of teeth and pulps play important roles in accurate classification. After using horizontal integral projection is first applied to separate the upper and lower jaw followed by vertical integral projection to each jaw. This paper [7] only classification is performed.

    In [2011], C.K. Modi in a proposed feature extraction technique for dental X-

    ray images based on multiple features [10]. In system include a good quality enhancement method then after radiograph segmentation. Feature extraction is used prior to matching the AM and PM images. The contour is extracted by Fourier descriptors are powerful for two-dimensional shape description and other combination method of gray level co-occurrence matrix (GLCM). Matching is done using both Euclidean distance and Hausdorff distance. It is done by finding the mean square error (MSE) between the query and database images. The retrival accuracy of multiple features (shape and texture) using Mean Square Error and Euclidean distance is 40%.All the methods discussed above are summarized in table 1.

  3. Conclusion & Future Work

    From the review of above papers, the main challenge in developing an automated dental identification system is to deal with poor quality of images, teeth overlap, imaging angle, teeth shape change consideration due to aging, etc. Then after matching of AM and PM images in dental radiograph. From the performance index found in the previous section, it is clear that the algorithm suggested by [2007], O. Nomir and M. Mottaleb by best using techniques among other techniques. Here one works to find a fast and better novel approach to enhance and segmentation method for dental radiograph and thereafter matching of PM image with AM images for better similarity results.


    1. Anil K. Jain, Hong Chen, Tooth contour extraction for matching dental radiographs, Pattern Recognition (2004) 1051 – 4651.

    2. Mohamed Abdel-Mottaleb, Omaima Nomir, Diaa Eldin Nassar , Gamal Fahmy, and Hany H. Ammar, Challenges of Developing an Automated Dental Identification System, 2004 IEEE.

    3. Hong chen and Anil K. Jain, Dental Biometrics: Alignment and Matching of Dental Radiographs,

      IEEE Transactions on Pattern Analysis and Machine Intelligence, Vol-27, Issue:8, Aug 2005.

    4. Omaima Nomir and Mohamed Abdel-Mottaleb, HIERARCHICAL DENTAL X-RAY RADIOGRAPHS MATCHING, IEEE 2006.

    5. Omaima Nomir and Mohamed Abdel-Mottaleb, Human Identification From Dental X-Ray Images Based on the Shape and Appearance of the Teeth, IEEE Transactions on Information Forensics and Security Vol-2, Issue:2, June 2007.

    6. Supaporn Kiattisin, Adisorn Leelasantitham, Kosin Chamnongthai and Kohji Higuchi, A Match of X-ray Teeth Films Using Image Processing Based on Special Features of Teeth, SICE Annual Conference, 20-22, Aug 2008.

    7. P. L. Lin, Y. H. Lai, An Effective Classification System for Dental Bitewing Radiographs Using Entire Tooth, Global Congress on Intelligent Systems, IEEE 2009.

    8. Samir Shah, Ayman Abaza, Arun Ross and Hany Ammar, Automatic Tooth Segmentation using Active Contour without edges, Biometrics Symposium,IEEE 2006.

    9. P. Choorat, W. Chiracharit and K. Chamnongthai, A Single Tooth Segmentation Using Structural Orientations and Statistical Textures, 2011 Biomedical Engineering International Conference, IEEE-2011.

    10. C. K. Modi, A Proposed Feature Extraction Technique for Dental X-Ray Images Based on Multiple Features,2011 International Conference on Communication Systems and Network Technologies, IEEE-2011.

    11. A. Ross, H. Ammar, Retrieving Dental Radiographs For Post-Mortem Identification, Appeared in Proc. Of IEEE International Conference on Image Processing (ICIP), Nov 2009.

[12]D. E.Nassar, A. Abaza, H. Ammar, Automatic Construction of Dental Charts for Post-mortem Identication, IEEE TRANSACTION ON INFORMATION FORENSICS AND SECURITY, VOL.3, No. 2, June,2008.

[13] S. Dighe, R. Shriram, Preprocessing , Segmentation and Matching of Dental Radiographs used in Dental Biometrics,ISSN No. 2278-3083, Vol 1, No. 2, May-June 2.






Feature &






Binary image masking

Iterative & Adaptive Thresholding; Horizontal & Vertical integral projection

Signature vector from tooth contour

Absolute distance matching

Out of 29 PM query images, Rank I-21, 4 out of 5PM images correctly


Anil K. Jain

Assumed good quality images were used

Horizontal & Vertical integral projection

Contour of Teeth & Shapes of Dental work

Computation of image distances matching

Out of 11 PM query images,Rank-I- 72%,Rank-IV-

91%, Rank-VII-


O. Nomir & M. Mottaleb

Binary image masking

Radiograph Segmentation: Iterative & Adaptive thresolding;

Horizontal & Vertical integral projection

Tooth Contour

Hierarchical Chamfer Matching Algorithm

Out of 50 PM query images, Rank I -40,Rank II-3, Rank III-2,

Rank V-4, Rank VII-1


S. Kiattisin

Brightness Adjustment; Binary image conversion

Chain code Decoding

Special features of teeth

Absolute matching between decoding &

statistical code

Same code match=90 %( same pattern); 50

%( different


P. L. Lin &

Y. H. Lai

Homomorphic Filtering;Adapti ve contrast stretching & Adaptive morphological


Thresholding; Horizontal & Vertical integral projection

Relative length/width ratio of a teeth, Relative crown size

Only classification is performed

94.9% for molars

& 95.6% for premolars.

S. Shah &

A. Ross

Assumed good quality images were used

Global(Corner Detection)& Local (Teeth Isolation)


Teeth Contour

Only teeth contour is performed

Contour extraction procedure is extremely fast app.


C. K. Modi

Assumed good quality images were used

Radiograph Segmentation

Fourier Descriptors & Gray level co-occurrence matrix(Shape

& Texture)

Absolute & Euclidean distance matching

66.667% precision using Euclidean distance & Mean square error.

Table 1.Dental Matching Technique Based On Human Identification

Leave a Reply