Validation of a Vision-Based, Three-Dimensional Facial Imaging System

A. AYOUB, PH.D., B.D.S., M.D.S., F.D.S., R.C.S.(ED.), F.D.S., R.C.P.S.(GLASG.) A. GARRAHY, B.A., M.B., B.CH., B.A.O., B.D.S., F.D.S., R.C.S.(ENG.) C. HOOD, B.D.S., F.D.S., R.C.P.S.(GLASG.), M.SC.D. J. WHITE, B.D.S., F.D.S., R.C.P.S.(GLASG.), M.ORTH., R.C.S.(ED.) M. BOCK, B.SC., B.MATH., PH.D. J.P. SIEBERT, B.SC., PH.D., M.I.E.E., C.ENG. R. SPENCER, B. ENG. A. RAY, M.B.B.S., M.S., F.R.C.S.(PLASTIC)
Objective: The aim of this study was to assess the accuracy of a newly developed three-dimensional (3D) imaging system in recording facial morphology. Methods: Twenty-one infants with cleft lip each had a full-face alginate impression taken at the time of primary lip repair, and a stone cast was constructed from each impression. Five anthropometric points were marked on each cast. Each cast was digitized, and the 3D co-ordinates of the ve points were obtained using a co-ordinate measuring machine (CMM, Ferranti) of documented accuracy (9.53 m). Each cast was scanned in four positions using a computerized stereophotogrammetry (C3D) system. The ve points were located on the 3D images, and their 3D co-ordinates were extracted by three operators. The co-ordinate systems produced by C3D were aligned, via translation and rotation, to match the CMM co-ordinate system using partial ordinary procrustes analysis. The displacements of the adjusted C3D co-ordinates from the reference co-ordinates were then measured. Three different types of errors were identied: operator, system, and registration errors. Results: Operator error was within 0.2 mm of the true co-ordinates of the landmarks. C3D was accurate within 0.4 mm. The average displacement of points over the 21 casts at four positions for the three operators was 0.79 mm (median 0.68). Conclusions: The presented 3D imaging system is reliable in recording facial deformity and could be utilized in recording cleft deformities and measuring the changes following surgery
KEY WORDS: cleft lip and palate, facial analysis, facial deformities, morphometrics, 3D imaging

Three-dimensional (3D) measurement and characterization of facial surface anatomy are fundamental to the objective

Dr. Ayoub is a Professor in Oral and Maxillofacial Surgery, Dr. Garrahy is a Clinical Research Assistant, Dr. Hood is a Lecturer in Pediatric Dentistry, and Dr. White is a Lecturer in Orthodontics, Glasgow Dental Hospital and School, Glasgow, Scotland. Dr. Bock is a Lecturer in Statistics, Department of Statistics, and Dr. Siebert is a Senior Research Fellow and Leader of the Vision and Graphics Laboratory (3D-MATIC), Department of Computing Science, University of Glasgow, Glasgow, Scotland. Mr. Spencer is with the School of Manufacturing and Mechanical Engineering, University of Birmingham, Birmingham, England. Mr. Ray is a Consultant in Plastic Surgery, Canniesburn Hospital, Glasgow, Scotland. This research was presented at the 9th International Congress on Cleft Palate and Related Craniofacial Anomalies; Goteborg, Sweden; 2529 June 2001. This project has been funded by a grant from the Chief Scientist Ofce in Scotland and the National Lottery Charities Board, the United Kingdom and supported by the UK Imaging Faraday Partnership. Submitted for publication June 2002; Accepted September 2002. Address correspondence to: Prof. Ashraf F. Ayoub, Oral and Maxillofacial Surgery, Leader of Biotechnology and Craniofacial Research Section (BACS), Glasgow Dental Hospital and School, 378 Sauchiehall Street, Glasgow G2 3JZ, Scotland, the UK. E-mail a.ayoub@dental.gla.ac.uk. 523

analysis of facial deformity, which is a 3D problem. Orthodontist and dentofacial surgeons deal with the physical relationships among the parts of the human head. These 3D physical relationships change through the course of growth, development, and treatment. Therefore, analysis of the results of corrective surgery in three dimensions is required and ideally it should also be possible to visualize the expected results of planned surgery preoperatively. Various techniques of measuring facial morphology have been reported, including direct anthropometry (Peyton and Ritchie, 1936; Farkas et al., 1993), laser scanning systems (Cutting et al., 1988; Moss et al., 1989; Bush and Antonyshyn, 1996), facial plaster models (Mishima et al., 1996), moire stripes (Kawai et al., 1990; Chen and Iizuka, 1995), video recording (Morrant and Shaw, 1996), and a liquid crystal range nder (Yamada et al., 1999). Recent innovations in computerized stereophotogrammetry (C3D) provide a useful technique for 3D recording of the face (Ayoub et al., 1996, 1997, 1998, 2001; Bourne et al., 2001). In 50 milliseconds, monochrome and color stereo images are captured. The integration of these images produces a dimen-

524

Cleft PalateCraniofacial Journal, September 2003, Vol. 40 No. 5

FIGURE 1 A capture station containing a pair of monochrome digital cameras to capture a stereoimage and a color digital camera to capture the skin texture in addition to a speckle texture projection ash and white light ash.

sionally accurate 3D range model and a colored photo realistic overlay (Figs. 1 and 2). This camera is not available commercially, and it is currently used for research purposes. The aim of this study was to assess the accuracy of the C3D system in locating landmarks in comparison with machine measurements taken independently at another center using a co-ordinate measuring machine (Ferranti, Birmingham, U.K.). MATERIALS
AND

METHODS

This investigation was conducted on 21 infants with cleft lip. A full facial alginate impression was taken, under general anesthesia, at the time of cleft repair, and stone models were constructed. Five points were marked on each facial model. The ve landmarks were the right corner of mouth (mouth R), left corner of the mouth (mouth L), edge of left nostril (nose L), tip of nose, and edge of right nostril (nose R). These were digitized twice

using a Ferranti co-ordinate-measuring machine (CMM) at the School of Manufacturing and Mechanical Engineering, University of Birmingham, U.K. The heart of the CMM is its probe, used to sense the object to measure. Repeatability error was determined by the manufacturer to be 2 m within a volume, covered by the axis accuracy measurements. Scale error is a combination of the accuracy of the axis of the CMM and the accuracy of the probes. The volumetric accuracy measured by the manufacturer at the University of Birmingham was 7.5 m. Maximum touch probe inaccuracy was 0.00053 mm. The sum of the volumetric error of the CMM and maximum probe error was 9.53 m (Spencer et al., 1996). The C3D Imaging System The facial casts were scanned using a new 3D imaging system, which consists of two camera stations placed at each side

FIGURE 2 A three-dimensional image of an infant with cleft captured by the computerized stereophotogrammetry system. A. Range model. B. Photorealistic model. C. Wire frame model. FIGURE 3 A three-dimensional model of one of the captured facial casts. A. Wire frame model. B. Photorealistic model. C. Range model.

Ayoub et al., 3D IMAGING OF THE FACE

525

526

Cleft PalateCraniofacial Journal, September 2003, Vol. 40 No. 5

of the face to take a stereo image. Each station contains a pair of monochrome digital cameras to capture a stereo image and a color digital camera to capture the skin texture (Fig. 1). The face is illuminated with a speckled ash that projects a random texture pattern onto the face. This textured illumination provides sufcient information in the images, captured by the monochrome cameras, to match the two sides of the face and accurately construct the 3D facial model. Ten milliseconds are required for exposure of the monochrome and color cameras, with a 30-millisecond time lapse between the two captures. It takes 50 milliseconds to capture the full face using the two camera stations. The resolution of the cameras is 766 by 576 usable pixels, with a focal length of 50 mm. A personal computer requires less than 5 minutes to produce a 3D model of the captured facial casts (Fig. 3). Scanning of the Facial Casts

Capture Error This was carried out by redigitizing the duplicate, sameposition, 3D images of six casts. A comparison of the discrepancies observed among models with those from stage 1 gave some indication of the magnitude of the combined operator and system error. Registration Error The third stage introduced the potential for registration error in landmark identication by comparing images of the same cast at different positions. Clearly the overall error also incorporates both operator and system error, and therefore this stage was the most complete, using images from all 21 casts taken at each of four positions. C3D System Errors

Each cast was positioned 50 cm in front of the camera and scanned in each of the following four positions in relation to the camera stations to analyze the possible effect of head position on the accuracy of the captured 3D image: (1) facing the center of the camera conguration (center); (2) rotated counterclockwise 20 degrees; (3) rotated clockwise 20 degrees; and (4) moved 10 cm closer and facing the center of the camera conguration (front). A single image of each cast in each position was built for digitization of landmarks. Eighty-four images were produced (21 casts 4 positions), and 420 landmarks (84 images 5 landmarks) were digitized. Digitization of Landmarks The landmarks were digitized on the computer screen by three observers, and the x, y, and z co-ordinates were recorded. A zoom-in facility was provided to increase accuracy in digitizing landmarks. To assess intraobserver reproducibility, each observer redigitized six randomly selected 3D images, generating 30 landmarks to analyze for each observer (6 casts 5 landmarks 30). To assess the reproducibility of the 3D scanning procedure, duplicate images of six randomly selected casts in a randomly selected position were redigitized by each observer, generating 3D co-ordinates of 30 landmarks for analysis (6 casts 5 landmarks 30). Data Analysis Three potential sources of error in landmark identication were analyzed at the following stages: Operator Error The rst stage involved redigitization, on two different occasions, of six randomly selected 3D images. This permitted the quantication of the operators placement accuracy.

Using ordinary partial procrustes analysis (OPPA; Dryden and Mardia, 1998), the 3D global positions of each landmark produced by C3D were aligned via translation and rotation to match the 3D global positions of the same landmarks that were produced by the CMM (reference gold standard). The null hypothesis was that there is a perfect match between the 3D global position of the corresponding landmarks that were generated by the two systems. The distance between each corresponding landmark, after superimposition, would represent the discrepancies between the two methods of measuring 3D coordinates of the facial landmarks. RESULTS Operator Error: Discrepancies Because of Inconsistencies When Locating Landmarks by Hand Because all of the data used to quantify the system and registration errors also contained operator error, it was therefore necessary to quantify the operator error as a rst step. To quantify the amount of error associated with placing landmarks, each image was marked up twice by each of three operators. Differences between repeatedly placed landmarks were calculated, and these values were averaged over the six images (for each combination of landmark and operator separately). From Table 1 it is clear that the magnitude of operator error is negligible for all practical considerations because repeatedly located landmarks were, on average, within 0.20 mm of each other with a minimum and maximum landmark/operator average of 0.04 mm and 0.57 mm, respectively. Capture Error: Discrepancies Because of Errors in the Underlying Model Generated by C3D Comparing the landmark congurations from different images of the same cast in the same position would quantify instability in the system. Table 2 shows that for all landmarks, facial casts,

Ayoub et al., 3D IMAGING OF THE FACE

527

TABLE 1 Average Displacement (in mm) of Repeatedly Placed Landmarks

Operator Landmarks 1 2 3 All

TABLE 2 Repeated Digitization of the Duplicate 3D Images (in mm) at the Same Position by Three Operators
Operator Landmark names 1 2 3 Grand Total

Mouth L Mouth R Nose L Nose R Tip Mean

0.160 0.120 0.141 0.291 0.206 0.179

0.182 0.137 0.079 0.153 0.039 0.124

0.239 0.279 0.263 0.571 0.189 0.301

0.191 0.179 0.161 0.322 0.145 0.199

Mouth L Mouth R Nose L Nose R Tip Grand Total

0.356 0.352 0.333 0.932 0.393 0.442

0.268 0.363 0.499 0.104 0.290 0.323

0.432 0.556 0.679 1.499 0.340 0.677

0.347 0.428 0.516 0.910 0.336 0.483

and operators observations, the average discrepancy was 0.48 mm (range 0.1 mm to 1.5 mm). Taking into consideration the magnitude of observers errors summarized in Table 1, this demonstrated a negligible inherent instability in the C3D system when multiple images of the same object are taken. Registration Error: Discrepancies Because of Differences in the Placement of the Imaged Object in Relation to the Cameras Registration error was estimated by placing the casts in four different positions relative to the C3D camera stations. Because the images to be compared are located in different places in the co-ordinate system, the congurations of landmarks derived from each image must be registered. OPPA that rotates and translates the co-ordinates of the C3D congurations to maximum superimposition was applied. The co-ordinate systems were all calibrated in millimeters. To assess the effect of each position on the digitized landmark co-ordinates, a series of pair-wise comparisons were made of mark-ups at different positions. These generated six comparisons for each of 21 casts for each of the three operators. Table 3 summarizes the results and shows the average within-landmark discrepancy was 0.42 mm (range 0.31 mm to 0.58 mm). Comparisons Between C3D Congurations and Birmingham Co-Ordinate Values The last set of results compared the co-ordinates of the C3D images with the landmark CMM. To compare the two sets of measurements, it was necessary to standardize the co-ordinate systems of each, and once again OPPA was used. Once the two co-ordinate systems have been registered, differences between the C3D and CMM co-ordinate values were of interest. Table 4 summarizes the distributions of 1260 values generated from images of 21 casts taken at four positions, each with ve landmarks and marked up by three operators. Rotating the facial cast 20 degrees to the right or placing the casts in the front position produced the least discrepancy in comparison with the CMM-produced co-ordinates. The landmarks, tip of the nose and nose R, showed the highest degree of difference. The averaged difference between C3D and CMM was 0.83 mm (range 0.41 mm to 1.52 mm; Table 4). When each position

of the facial cast was considered, analysis of variance (ANOVA) showed that the landmarks, mouth L and mouth R, had signicantly smaller discrepancies on average than those associated with the rest of the landmarks. Comparisons of the Different Types of Errors The validation study was designed so that the four potential types of errors (operator, capture, registration, and system) could be investigated. By comparing the displacements observed among different pairs of congurations, insight can be gained as to the relative magnitude of different types of errors in the observed landmarks. Figure 4 displays box plots of data corresponding to the cumulative effects of the four types of errors. The ANOVA on the log transformation of these data conrmed the subjective impression that there are differences among the means of the cumulative errors (p .000). Pairwise comparisons using Students t tests on the transformed data are summarized in Table 5. These conrmed that the signicant differences lie between mean operator error and each of the other errors. There was neither signicant difference between the mean capture error and mean registration error nor evidence of a difference between system and registration errors. Note that this lack of a signicant difference between the registration and system errors can be accounted for by the presence of two outliers in the registration errors. In summary, the total error associated with identifying landmarks was signicantly greater than the error associated with either, repeatedly locating the same point on the same image or locating points on different images of the same cast in the same position. Although this result may appear alarming, the magnitude of the total error was, on average, less than 1 mm. Differences Among Observers (Interobserver Discrepancies) No statistically signicant differences between the reproducibility of landmark identication among the observers were found using ANOVA. DISCUSSION The system error of the C3D imaging system described above is comparable to that of other 3D imaging systems such

528

Cleft PalateCraniofacial Journal, September 2003, Vol. 40 No. 5

TABLE 3 The Average Within-Landmark Discrepancy (in mm) Caused by Different Positions of the Imaged Facial Casts in Relation to the Cameras With a Minimum Average of 0.31 mm and a Maximum Average of 0.58 mm
Center versus Front Center versus 20degR* Center versus 20degL* 20degR versus Front* 20degL versus Front* 20degL versus 20degR*

Landmarks

Grand Total

Mouth L Mouth R Nose L Nose R Tip Grand total

* 20degR

0.456 0.456 0.447 0.370 0.513 0.456

0.383 0.336 0.365 0.439 0.452 0.391

0.324 0.422 0.358 0.383 0.579 0.416

rotated clockwise 20 degrees.

0.334 0.312 0.369 0.410 0.422 0.365

0.366 0.575 0.434 0.390 0.538 0.465

0.311 0.350 0.521 0.443 0.485 0.422

0.364 0.414 0.415 0.398 0.499 0.421

rotated counterclockwise 20 degrees; 20degL

as laser scanning and video imaging (Lowey, 1993; Moss et al., 1994; Ferrario et al., 1997; Trotman et al., 1997). The speed of the data collection process allows the use of C3D in infants and children because minimal cooperation is required. This investigation was conducted on plaster casts of the face, only to provide a 3D object that could be marked and measured by both the CMM and the newly developed C3D imaging system. Recording the face with an alginate impression would distort the facial soft tissue, and our aim was to utilize the 3D camera system to capture the face directly in patients with cleft lip and palate. However, without assessing the magnitude of errors associated with using the 3D cameras on inanimate objects, we could not proceed to an analysis of the possible errors of recording the face directly. This investigation highlighted three different types of errors that could arise from digitizing 3D images captured by 3D cameras. The reproducibility of digitizing landmarks, operator error, was signicantly low and was within 0.2 mm from the true co-ordinates of the landmarks. C3D was accurate in capturing facial models in that the average discrepancies of the landmarks co-ordinates in comparison with those recorded using the CMM was 0.4 mm. This also included the operator error, which constituted less than 50% of this error. The results of this investigation highlighted that the total error associated with digitizing landmarks on 3D images is signicantly greater than the error associated with either repeatedly locating the same points on the same images or locating landmarks on the different images of the same facial cast in the same position. However, the magnitude of the total error on average was less than 1 mm. With regard to the landmarks that were digitized on the 3D models, some of these were more readily identied than others.
TABLE 4 Average Distances (in mm) Among Landmarks as a Result of the discrepancy Between C3D and CMM Coordinate Values*
Landmarks 20degL 20degR Center Front Grand Total

The nostril right proved to be difcult to digitize, possibly because of inadequate lighting at the base of the nose. Identication of other landmark sets on the cheek, lips, and chin may produce a greater error; this would require further investigation. It was interesting to investigate whether positioning of the facial cast relative to the C3D camera kit would affect the accuracy of extracting landmark co-ordinates. There were statistically signicant differences between positions with the close front and rotation of the cast 20 degrees to the right producing the most accurate data. The close front position would provide a close-up view of the object to be scanned and equal distribution of lighting on both sides. There is no clear reason that 20-degree rotation to the right would produce more accurate results than rotation of the object 20 degrees to the left. The only explanation is that most of the facial casts were for left side cleft cases. Rotation to the right would increase illumination of the left side, improving the quality of capture and the accuracy in digitizing landmarks on the left side of the face. This investigation conrms the reliability of our 3D imaging system in recording facial morphology. The technique implements a software architecture that efciently integrates stereoimage capture, stereoimage matching, photogrammetry, surface model construction, and visualization of this model. C3D affords a number of advantages over other noncontact optical measurement techniques. The geometric simplicity of the capture hardware and well-understood calibration methods employed ensure that the accuracy is superior to other more optically complex techniques (Urquhart, 1997). Perhaps more

Mouth L Mouth R Nose L Nose R Tip Grand total

0.63 0.59 0.87 1.52 1.46 1.02

0.54 0.45 0.75 0.79 0.73 0.63

0.63 0.55 0.77 1.249 1.20 0.86

0.64 0.41 0.86 0.99 1.11 0.79

0.61 0.50 0.81 1.18 1.14 0.83

* 20degL rotated clockwise 20 degrees; 20degR C3D computerized stereophotogrammetry; CMM

rotated counterclockwise 20 degrees; co-ordinate-measuring machine.

FIGURE 4 Box plots of the data corresponding to the cumulative effects of the four types of errors.

Ayoub et al., 3D IMAGING OF THE FACE

529

TABLE 5 Pairwise Comparisons of Different Types of Errors Showing No Signicant Difference Between the Mean Capture Error and Mean Registration Error and No Difference Between System and Registration Error
Pairwise Comparison Value

REFERENCES
Ayoub AF, Garrahy A, Hood C, White J, Bock M, Spencer R. Validation of a vision based three-dimensional facial imaging system. In: Transactions of the 9th International Congress on Cleft Palate and Related Craniofacial Anomalies. Gotenburg: Elanders Novum; 2001:283285. Ayoub AF, Siebert P, Moos K, Wray D, Urquhart C, Niblett T. A vision-based three-dimensional capture system for maxillofacial assessment and surgical planning. Brit J Oral Maxillofacial Surg. 1998;36:353357. Ayoub AF, Wray D, Moos KF, Siebert P, Jin J, Niblett T, Urquhart C, Mowforth P. A three-dimensional imaging system for archiving dental study model: a preliminary report. Int J Adult Orthod Orthognath Surg. 1997;12:7984. Ayoub AF, Wray D, Moos KF, Siebert P, Jin J, Niblett T, Urquhart C, Mowforth P. Three-dimensional modeling for modern diagnosis and planning in maxillofacial surgery. Int J Adult Orthod Orthognath Surg. 1996;11:225233. Bourne C, Kerr W, Ayoub A. Development of a three-dimensional imaging system for analysis of facial change. Clin Orthod Res. 2001;4:105111. Bush K, Antonyshyn O. Three-dimensional facial anthropometry using a laser surface scanner: validation of the technique. Plast Reconstr Surg. 1996;98: 226235. Chen LH, Iizuka T. Evaluation and prediction of the facial appearance after surgical correction of mandibular hyperplasia. Int J Oral Maxillofac Surg. 1995;24:322326. Cutting CB, McCarthy JG, Karron DB. Three-dimensional input of body surface data using a laser light scanner. Am Plast Surg. 1988;21:3846. Dryden IL, Mardia KV. Statistical Shape Analysis. Chichester: John Wiley and Sons; 1998;9295. Farkas LG, Hajnis K, Posnick JC. Anthropometric and anthroscopic ndings of the nasal and facial region in cleft patients before and after primary lip and palate repair. Cleft Palate Craniofac J. 1993;30:112. Ferrario VF, Sforza C, Poggio CE, Cova M, Tartaglia G. Preliminary evaluation of an electromagnetic three-dimensional digitiser in facial anthropometry. Cleft Palate Craniofac J. 1997;35:915. Kawai T, Natsume N, Shibata H, Yamamoto T. Three-dimensional analysis of facial morphology using moire stripes. Part I. Method. Int J Oral Maxillofac Surg. 1990;19:356358. Lowey M. The development of a new method of cephalometric and study cast measurements with computer controlled video image capture system. Brit J Orthod. 1993;20:203214. Mishima K, Sugahara T, Mori Y, Sakuda M. Application of a new method for anthropometric analysis of the nose. Plastic Reconstr Surg. 1996;637644. Morrant DG, Shaw WC. Use of standardised video recordings to assess cleft surgery outcome. Cleft Palate Craniofac J. 1996;33:134142. Moss JP, Linney AD, Grindrod SR, Mosse A. Laser scanning system for the measurement of facial surface morphology. Opt Laser Engl. 1989;10:179190. Moss JP, McCance AM, Fright WR, Linney AD, Janes DR. A three-dimensional soft tissue analysis of fteen patients with class II, division I malocclusion after bimaxillary surgery. Am J Orthod Dentofacial Orthop. 1994; 105:430437. Peyton WT, Ritchie HP. Quantitative studies on congenital clefts of the lip. Arch Surg. 1936;33:10461035. Siebert JP, Marshall SJ. Human body 3D imaging by speckle texture projection photogrammetry. Sens Rev. 2000;20:218226. Spencer R, Hathaway R, Speculand B. 3D computer data capture and imaging applied to the face and jaws. Br J Oral Maxillofac Surg. 1996;34:118123. Trotman CA, Stohler C, Johnston L. Measurement of Facial Soft Tissue Mobility in Man. Cleft Palate Craniofac J. 1997;35:1625. Urquhart CW. The Active Stereo Probe: The Design and Implementation of an Active Videometrics System. Glasgow: University of Glasgow, Department of Computing Science; 1997. Dissertation. Yamada T, Sugahara T, Mori Y, Minami K, Sakuda M. Development of a 3D measurement and evaluation system for facial forms with a liquid crystal range nder. Comput Methods Programs Biomed. 1999;58:159173.

Operator versus capture Operator versus registration Operator versus system Capture versus registration Capture versus system Registration versus system

.000 .000 .000 .544 .011 .081

importantly, the technique is also a full-eld one, producing 3D information from the whole scene without the need for scanning. This leads to an inherently faster and consequently more suitable data-capture method than nonfull eld techniques such as laser scanning triangulation. It also possesses an advantage over other full-eld techniques, such as phasestepping moire fringe contouring, temporal light modulation (structured light technique), in that only a single image is required per camera (Siebert and Marshall, 2000). The method fullls the following clinical requirements: 1. 2. 3. 4. 5. 6. 7. 8. Fast capture time of the face. Identication of facial landmarks to within 1 mm. Simplicity of the input operation. Three-dimensional display within a few minutes. Ease of integration into existing medical practice. Cost-effectiveness. Patient is not exposed to harmful radiation. Reliable technique for data storing and archiving.

The system has a wide range of clinical usage as well as providing a tool for clinical auditing by monitoring progressive outcomes throughout a course of treatment or following surgery. This technique is currently being applied to the study of facial morphology and growth in infants and preschool children with and without orofacial clefting in Scotland. It has also been used to study reproducibility of adult facial expressions and treatment outcome assessment in adults with orofacial clefting. In conclusion, this investigation conrmed the reproducibility and validity of the C3D method in recording facial morphology. The development of this noninvasive fast-capture technique enables the next step toward building a population norm for comparison with patients with cleft lip and palate. This will provide a unique opportunity for researchers to measure 3D facial growth in cleft and noncleft populations, quantify the magnitude of facial deformity, and assess changes following surgery.