3D Shape Model Research Articles

Building 3D Statistical Shape Models of Horticultural Products

A method to build a 3D statistical shape model of horticultural products is described. The framework consists of two parts. First, the surfaces of the horticultural products, which are extracted from X-ray CT scans, are registered to obtain meaningful correspondences between the surfaces. In the second part, a statistical shape model is built from these corresponded surfaces, which maps out the variability of the surfaces and allows to generate new, realistic surfaces. The proposed shape modelling method is applied to 30 Jonagold apples, 30 bell peppers, and 52 zucchini. The average geometric registration error between the original instance and the deformed reference instance is 0.015 ± 0.011 m m for the apple dataset, 0.106 ± 0.026 m m for the bell pepper dataset, and 0.027 ± 0.007 m m for the Zucchini dataset. All shape models are shown to be an excellent representation of their specific population, as they are compact and able to generalize to an unseen sample of the population.

Study of the interactions between proximal femur 3d bone shape, cartilage health, and biomechanics in patients with hip Osteoarthritis.

In this study quantitative MRI and gait analysis were used to investigate the relationships between proximal femur 3D bone shape, cartilage morphology, cartilage biochemical composition, and joint biomechanics in subject with hip Osteoarthritis (OA). Eighty subjects underwent unilateral hip MR-imaging: T1ρ and T2 relaxation times were extracted through voxel based relaxometry and bone shape was assessed with 3D MRI-based statistical shape modeling. In addition, 3D gait analysis was performed in seventy-six of the studied subjects. Associations between shape, cartilage lesion presence, severity, and cartilage T1ρ and T2 were analyzed with linear regression and statistical parametric mapping. An ad hoc analysis was performed to investigate biomechanics and shape associations. Our results showed that subjects with a higher neck shaft angle in the coronal plane (higher mode 1, coxa valga), thicker femoral neck and a less spherical femoral head (higher mode 5, pistol grip) exhibited more severe acetabular and femoral cartilage abnormalities, showing different interactions with demographics factors. Subjects with coxa valga also demonstrated a prolongation of T1ρ and T2. Subjects with pistol grip deformity exhibited reduced hip internal rotation angles and subjects with coxa valga exhibited higher peak hip adduction moment and moment impulse. The results of this study establish a clear relationship between 3D proximal femur shape variations and markers of hip joint degeneration-morphological, compositional, well as insight on the possible interactions with demographics and biomechanics, suggesting that 3D MRI-based bone shape maybe a promising biomarker of early hip joint degeneration. © 2017 Orthopaedic Research Society. Published by Wiley Periodicals, Inc. J Orthop Res 36:330-341, 2018.

3D shape of asteroid (6) Hebe from VLT/SPHERE imaging: Implications for the origin of ordinary H chondrites

Context. The high-angular-resolution capability of the new-generation ground-based adaptive-optics camera SPHERE at ESO VLT allows us to assess, for the very first time, the cratering record of medium-sized (D~100-200 km) asteroids from the ground, opening the prospect of a new era of investigation of the asteroid belt's collisional history. Aims. We investigate here the collisional history of asteroid (6) Hebe and challenge the idea that Hebe may be the parent body of ordinary H chondrites, the most common type of meteorites found on Earth (~34% of the falls). Methods. We observed Hebe with SPHERE as part of the science verification of the instrument. Combined with earlier adaptive-optics images and optical light curves, we model the spin and three-dimensional (3D) shape of Hebe and check the consistency of the derived model against available stellar occultations and thermal measurements. Results. Our 3D shape model fits the images with sub-pixel residuals and the light curves to 0.02 mag. The rotation period (7.274 47 h), spin (343 deg,+47 deg), and volume-equivalent diameter (193 +/- 6km) are consistent with previous determinations and thermophysical modeling. Hebe's inferred density is 3.48 +/- 0.64 g.cm-3 , in agreement with an intact interior based on its H-chondrite composition. Using the 3D shape model to derive the volume of the largest depression (likely impact crater), it appears that the latter is significantly smaller than the total volume of close-by S-type H-chondrite-like asteroid families. Conclusions. Our results imply that (6) Hebe is not the most likely source of H chondrites. Over the coming years, our team will collect similar high-precision shape measurements with VLT/SPHERE for ~40 asteroids covering the main compositional classes, thus providing an unprecedented dataset to investigate the origin and collisional evolution of the asteroid belt.

Spin states of asteroids in the Eos collisional family

Eos family was created during a catastrophic impact about 1.3 Gyr ago. Rotation states of individual family members contain information about the history of the whole population. We aim to increase the number of asteroid shape models and rotation states within the Eos collision family, as well as to revise previously published shape models from the literature. Such results can be used to constrain theoretical collisional and evolution models of the family, or to estimate other physical parameters by a thermophysical modeling of the thermal infrared data. We use all available disk-integrated optical data (i.e., classical dense-in-time photometry obtained from public databases and through a large collaboration network as well as sparse-in-time individual measurements from a few sky surveys) as input for the convex inversion method, and derive 3D shape models of asteroids together with their rotation periods and orientations of rotation axes. We present updated shape models for 15 asteroids and new shape model determinations for 16 asteroids. Together with the already published models from the publicly available DAMIT database, we compiled a sample of 56 Eos family members with known shape models that we used in our analysis of physical properties within the family. Rotation states of asteroids smaller than ∼ 20 km are heavily influenced by the YORP effect, whilst the large objects more or less retained their rotation state properties since the family creation. Moreover, we also present a shape model and bulk density of asteroid (423) Diotima, an interloper in the Eos family, based on the disk-resolved data obtained by the Near InfraRed Camera (Nirc2) mounted on the W.M. Keck II telescope.

Product sizing with 3D anthropometry and [formula omitted]-medoids clustering

Aside from anthropometric data tables, 3D shape models of the human body are becoming increasingly common and call for new product sizing methods based on 3D anthropometry. Though some shape model-based methods exist, most of them focus on mathematical clustering and do not discuss the usability of the clustering results for product design. In this paper, a new shape-model based clustering method for product sizing is presented that takes into account both shape information and usability for designers. The new method, called constrained k-medoids clustering, is applied on a shape model of 100 human heads. It is compared to a partitioning around medoid (PAM) clustering of anthropometric measurements of the same 100 heads (i.e., feature-based), as well as to PAM clustering of the shape model (i.e., shape based). Results show that both shape-based and constrained clustering perform better than feature-based clustering, with an average size-weighted variance (SWV) of 62×103±16×103 and 66×103±26×103 as compared to 72×103±12×103, respectively. The average point-to-point distances in shape-based and constrained k-medoids were found to be similar to those of feature-based k-medoids, indicating that using 3D-anthropometry for product sizing will not have a negative impact on designer workload and/or a higher cost to implement more sizes. The results suggest that for head-based products, which require accurate shape and size fit, sizing systems should be created using either shape-based or constrained k-medoids, with the latter being slightly less accurate but more intuitive for further design and verification.

Super-vector coding features extracted from both depth buffer and view-normal-angle images for part-based 3D shape retrieval

We have witnessed 3D shape models abundant in many application fields including 3D CAD/CAM, augmented/mixed reality (AR/MR), and entertainment. Creating 3D shape models from scratch is still very expensive. Efficient and accurate methods for shape retrieval is essential for 3D shape models to be reused. To retrieve similar 3D shape models, one must provide an arbitrary 3D shape as a query. Most of the research on 3D shape retrieval has been conducted with a “whole” shape as a query (aka whole-to-whole shape retrieval), while a “part” shape (aka part-to-whole shape retrieval) is more practically requested as a query especially by mechanical engineering with 3D CAD/CAM applications. A “part” shape is naturally constructed by a 3D range scanner as an input device. In this paper, we focus on the efficient method for part-to-whole shape retrieval where the “part” shape is assumed to be given by a 3D range scanner. Specifically, we propose a Super-Vector coding feature with SURF local features extracted from the View-Normal-Angle image, or the image synthesized by taking account of the angle between the view vector and the surface normal vector, together with the depth-buffered image, for part-to-whole shape retrieval. In addition, we propose a weighted whole-to-whole re-ranking method taking advantage of global information based on the result of part-to-whole shape retrieval. Through experiments we demonstrate that our proposed method outperforms the previous methods with or without re-ranking.

Analysis of the three-dimensional anatomical variance of the distal radius using 3D shape models

BackgroundVarious medical fields rely on detailed anatomical knowledge of the distal radius. Current studies are limited to two-dimensional analysis and biased by varying measurement locations. The aims were to 1) generate 3D shape models of the distal radius and investigate variations in the 3D shape, 2) generate and assess morphometrics in standardized cut planes, and 3) test the model’s classification accuracy.MethodsThe local radiographic database was screened for CT-scans of intact radii. 1) The data sets were segmented and 3D surface models generated. Statistical 3D shape models were computed (overall, gender and side separate) and the 3D shape variation assessed by evaluating the number of modes. 2) Anatomical landmarks were assigned and used to define three standardized cross-sectional cut planes perpendicular to the main axis. Cut planes were generated for the mean shape models and each individual radius. For each cut plane, the following morphometric parameters were calculated and compared: maximum width and depth, perimeter and area. 3) The overall shape model was utilized to evaluate the predictive value (leave one out cross validation) for gender and side identification within the study population.ResultsEighty-six radii (45 left, 44% female, 40 ± 18 years) were included. 1) Overall, side and gender specific statistical 3D models were successfully generated. The first mode explained 37% of the overall variance. Left radii had a higher shape variance (number of modes: 20 female / 23 male) compared to right radii (number of modes: 6 female / 6 male). 2) Standardized cut planes could be defined using anatomical landmarks. All morphometric parameters decreased from distal to proximal. Male radii were larger than female radii with no significant side difference. 3) The overall shape model had a combined median classification probability for side and gender of 80%.ConclusionsStatistical 3D shape models of the distal radius can be generated using clinical CT-data sets. These models can be used to assess overall bone variance, define and analyze standardized cut-planes, and identify the gender of an unknown sample. These data highlight the potential of shape models to assess the 3D anatomy and anatomical variance of human bones.

Hessian-constrained detail-preserving 3D implicit reconstruction from raw volumetric dataset

Massive routinely-acquired raw volumetric datasets are hard to be deeply exploited by downstream applications due to the challenges in accurate and efficient shape modeling. This paper systematically advocates an interactive 3D shape modeling framework for raw volumetric datasets by iteratively optimizing Hessian-constrained local implicit surfaces. The key idea is to incorporate contour based interactive segmentation into the generalized local implicit surface reconstruction. Our framework allows a user to flexibly define derivative constraints up to the second order via intuitively placing contours on the cross sections of volumetric images and fine-tuning the eigenvector frame of Hessian matrix. It enables detail-preserving local implicit representation while combating certain difficulties due to ambiguous image regions, low-quality irregular data, close sheets, and massive coefficients involved extra computing burden. To this end, we propose several novel technical elements, including data-specific importance sampling for adaptive spherical-cover generation, close sheet determination based on distinguishable local samples, and parallel acceleration for local least squares fitting. Moreover, we conduct extensive experiments on some volumetric images with blurry object boundaries, and make comprehensive, quantitative performance evaluation between our method and the state-of-the-art radial basis function based techniques. And we also apply our method to two practical applications. All the results demonstrate our method’s advantages in the accuracy, detail-preserving, efficiency, and versatility of shape modeling.

Machine Learning for Large-Scale Quality Control of 3D Shape Models in Neuroimaging.

As very large studies of complex neuroimaging phenotypes become more common, human quality assessment of MRI-derived data remains one of the last major bottlenecks. Few attempts have so far been made to address this issue with machine learning. In this work, we optimize predictive models of quality for meshes representing deep brain structure shapes. We use standard vertex-wise and global shape features computed homologously across 19 cohorts and over 7500 human-rated subjects, training kernelized Support Vector Machine and Gradient Boosted Decision Trees classifiers to detect meshes of failing quality. Our models generalize across datasets and diseases, reducing human workload by 30-70%, or equivalently hundreds of human rater hours for datasets of comparable size, with recall rates approaching inter-rater reliability.

Single image 3D human pose estimation using a procrustean normal distribution mixture model and model transformation

3D human pose estimation from a single image is an important problem in computer vision with a number of applications, including action recognition and scene understanding. However, it is still challenging due to its ill-posedness and complex non-rigid shape variations of a human body. In this paper, we use the Procrustean normal distribution mixture model as a 3D shape prior and propose a model transformation method for adjusting limb lengths of the 3D shape prior model, by which the proposed method can be applied to a novel test image. Inaccuracies of 2D part detections are handled by selecting from a diverse set of 2D pose candidates considering both the 2D part model and 3D shape model. Experimental results show that the proposed method performs favorably compared with existing methods, despite inaccuracies of 2D part detections and 3D shape ambiguities.

Analysis of the articular cartilage T1ρ and T2 relaxation times changes after ACL reconstruction in injured and contralateral knees and relationships with bone shape.

The objectives of this study were twofold: (1) to evaluate the longitudinal change in cartilage T1ρ and T2 6- and 12-months after ACL reconstruction (ACLR) in both reconstructed and intact contralateral knees with the aim of validating the role of the contralateral knee as an internal control in longitudinal studies; (2) to explore relationships between bone shape at the time of injury and the progression of T1ρ and T2 over 12-months after ACLR. T1ρ and T2 cartilage relaxation times and 3D MRI-based statistical shape modeling (SSM) of tibia and femur were computed for both knees of forty ACL-injured patients and 15 healthy controls. ACL subjects were scanned 8.4 ± 6.4 weeks after injury (2.4 ± 3.7 weeks prior to ACLR), 6- and 12-months after ACLR. Longitudinal changes in T1ρ and T2 values were assessed using linear mixed model, and partial correlation coefficients were calculated between bone shape and longitudinal changes in T1ρ and T2 values. Significant longitudinal increases in T1ρ and T2 values were observed in reconstructed and contralateral knees 6-months after ACLR. Tibial bone shape features, associated with the medial plateau height and width, were observed to be correlated with cartilage T1ρ and T2 progression in reconstructed knees. Our results suggest that caution should be used in considering contralateral knee as internal controls in longitudinal ACL studies and 3D MRI-based-SSM might serve as an imaging biomarker for the early stratification of patients at risk for developing post-traumatic accelerated cartilage degeneration and potentially osteoarthritis after ACL tear. © 2016 Orthopaedic Research Society. Published by Wiley Periodicals, Inc. J Orthop Res 35:707-717, 2017.

Automatic segmentation of the glenohumeral cartilages from magnetic resonance images

Magnetic resonance (MR) imaging plays a key role in investigating early degenerative disorders and traumatic injuries of the glenohumeral cartilages. Subtle morphometric and biochemical changes of potential relevance to clinical diagnosis, treatment planning, and evaluation can be assessed from measurements derived from in vivo MR segmentation of the cartilages. However, segmentation of the glenohumeral cartilages, using approaches spanning manual to automated methods, is technically challenging, due to their thin, curved structure and overlapping intensities of surrounding tissues. Automatic segmentation of the glenohumeral cartilages from MR imaging is not at the same level compared to the weight-bearing knee and hip joint cartilages despite the potential applications with respect to clinical investigation of shoulder disorders. In this work, the authors present a fully automated segmentation method for the glenohumeral cartilages using MR images of healthy shoulders. The method involves automated segmentation of the humerus and scapula bones using 3D active shape models, the extraction of the expected bone-cartilage interface, and cartilage segmentation using a graph-based method. The cartilage segmentation uses localization, patient specific tissue estimation, and a model of the cartilage thickness variation. The accuracy of this method was experimentally validated using a leave-one-out scheme on a database of MR images acquired from 44 asymptomatic subjects with a true fast imaging with steady state precession sequence on a 3 T scanner (Siemens Trio) using a dedicated shoulder coil. The automated results were compared to manual segmentations from two experts (an experienced radiographer and an experienced musculoskeletal anatomist) using the Dice similarity coefficient (DSC) and mean absolute surface distance (MASD) metrics. Accurate and precise bone segmentations were achieved with mean DSC of 0.98 and 0.93 for the humeral head and glenoid fossa, respectively. Mean DSC scores of 0.74 and 0.72 were obtained for the humeral and glenoid cartilage volumes, respectively. The manual interobserver reliability evaluated by DSC was 0.80 ± 0.03 and 0.76 ± 0.04 for the two cartilages, implying that the automated results were within an acceptable 10% difference. The MASD between the automatic and the corresponding manual cartilage segmentations was less than 0.4 mm (previous studies reported mean cartilage thickness of 1.3 mm). This work shows the feasibility of volumetric segmentation and separation of the glenohumeral cartilages from MR images. To their knowledge, this is the first fully automated algorithm for volumetric segmentation of the individual glenohumeral cartilages from MR images. The approach was validated against manual segmentations from experienced analysts. In future work, the approach will be validated on imaging datasets acquired with various MR contrasts in patients.

Sparse Representation for 3D Shape Estimation: A Convex Relaxation Approach.

We investigate the problem of estimating the 3D shape of an object defined by a set of 3D landmarks, given their 2D correspondences in a single image. A successful approach to alleviating the reconstruction ambiguity is the 3D deformable shape model and a sparse representation is often used to capture complex shape variability. But the model inference is still challenging due to the nonconvexity in the joint optimization of shape and viewpoint. In contrast to prior work that relies on an alternating scheme whose solution depends on initialization, we propose a convex approach to addressing this challenge and develop an efficient algorithm to solve the proposed convex program. We further propose a robust model to handle gross errors in the 2D correspondences. We demonstrate the exact recovery property of the proposed method, the advantage compared to several nonconvex baselines and the applicability to recover 3D human poses and car models from single images.

Model-based generation of personalized full-body 3D avatars from uncalibrated multi-view photographs

According to a number of studies, the use of personalized avatars in virtual environments can enhance the immersion experience of users and the effectiveness of communication between different players. The benefits of using personalized avatars in conjunction with recent technological developments in Virtual Reality (VR) prompted an increasing demand for low-cost systems that are capable of fast and easy generation of personal avatars for use in VR applications. In this paper we present a novel model-based technique that is capable of generating personalized full-body 3D avatars from orthogonal photographs. The proposed method utilizes a statistical model of human 3D shape and a multi-view statistical 2D shape model of its corresponding silhouettes. Our technique is automatic, requiring minimal user intervention, and does not need a calibrated camera. Each component of our proposed technique is extensively evaluated and validated, in an attempt to test the geometric accuracy and identifiability of generated avatars. Furthermore, we demonstrate the use of the proposed method for generating and importing animation-ready avatars in collaborative VR environments.

Aswan site on comet 67P/Churyumov-Gerasimenko: Morphology, boulder evolution, and spectrophotometry

We provide a detailed morphological analysis of the Aswan site on comet 67P/Churyumov-Gerasimenko (67P). We derive the size-frequency distribution of boulders ≥2 m and correlate this distribution with the gravitational slopes for the first time on a comet. We perform the spectral analysis of this region to understand if possible surface variegation is related to thedifferent surface textures observable on the different units. Methods. We used two OSIRIS Narrow Angle Camera (NAC) image data sets acquired on September 19 and 22, 2014, with a scale of 0.5 m/px. Gravitational slopes derived from the 3D shape model of 67P were used to identify and interpret the different units of the site. By means of the high-resolution NAC data sets, boulders ≥2.0 m can be unambiguously identified and extracted using the software ArcGIS. Coregistered and photometrically corrected color cubes were used to perform the spectral analyses, and we retrieved the spectral properties of the Aswan units. Results. The high-resolution morphological map of the Aswan site (0.68 km2) shows that this site is characterized by four different units: fine-particle deposits located on layered terrains, gravitational accumulation deposits, taluses, and the outcropping layered terrain. Multiple lineaments are identified on the Aswan cliff, such as fractures, exposed layered outcrops, niches, and terraces. Close to the terrace margin, several arched features observed in plan view suggest that the margin progressively retreats as a result of erosion. The size-frequency of boulders ≥2 m in the entire study area has a power-law index of −3.9 +0.2/−0.3 (1499 boulders ≥2 m/km2), suggesting that the Aswan site is mainly dominated by gravitational events triggered by sublimation and/or thermal insolation weathering causing regressive erosion. The boulder size-frequency distribution versus gravitational slopes indicates that when higher gravitational slope terrains are considered, only boulders ≤10 m are identified, as well as steeper power-slope indices. In addition, no boulders ≥2 m are observed on slopes ≥50°. This may indicate that larger blocks detached from a sublimating cliff cannot rest at these slopes and consequently fall down. The spectral analysis performed on the site shows that despite different morphologic units, no spectral differences appear in the multiple textures. This may confirm a redistribution of particles across the nucleus as a consequence of airfall, whether coming from Hapi or from the southern hemisphere when it is active during perihelion.

Ergonomic design of an EEG headset using 3D anthropometry

Although EEG experiments over the past decades have shown numerous applications for brain-computer interfacing (BCI), there is a need for user-friendly BCI devices that can be used in real-world situations. 3D anthropometry and statistical shape modeling have been shown to improve the fit of devices such as helmets and respirators, and thus they might also be suitable to design BCI headgear that better fits the size and shape variation of the human head. In this paper, a new design method for BCI devices is proposed and evaluated. A one-size-fits-all BCI headset frame is designed on the basis of three digital mannequins derived from a shape model of the human head. To verify the design, the geometric fit, stability and repeatability of the prototype were compared to an EEG cap and a commercial BCI headset in a preliminary experiment. Most design specifications were met, and all the results were found to be similar to those of the commercial headset. Therefore, the suggested design method is a feasible alternative to traditional anthropometric design for BCI headsets and similar headgear.

Estimating 3D Leaf and Stem Shape of Nursery Paprika Plants by a Novel Multi-Camera Photography System.

For plant breeding and growth monitoring, accurate measurements of plant structure parameters are very crucial. We have, therefore, developed a high efficiency Multi-Camera Photography (MCP) system combining Multi-View Stereovision (MVS) with the Structure from Motion (SfM) algorithm. In this paper, we measured six variables of nursery paprika plants and investigated the accuracy of 3D models reconstructed from photos taken by four lens types at four different positions. The results demonstrated that error between the estimated and measured values was small, and the root-mean-square errors (RMSE) for leaf width/length and stem height/diameter were 1.65 mm (R2 = 0.98) and 0.57 mm (R2 = 0.99), respectively. The accuracies of the 3D model reconstruction of leaf and stem by a 28-mm lens at the first and third camera positions were the highest, and the number of reconstructed fine-scale 3D model shape surfaces of leaf and stem is the most. The results confirmed the practicability of our new method for the reconstruction of fine-scale plant model and accurate estimation of the plant parameters. They also displayed that our system is a good system for capturing high-resolution 3D images of nursery plants with high efficiency.

Face reconstruction from image sequences for forensic face comparison

The authors explore the possibilities of a dense model-free three-dimensional (3D) face reconstruction method, based on image sequences from a single camera, to improve the current state of forensic face comparison. They propose a new model-free 3D reconstruction method for faces, based on the Lambertian reflectance model to estimate the albedo and to refine the 3D shape of the face. This method avoids any form of bias towards face models and is therefore suitable in a forensic face comparison process. The proposed method can reconstruct frontal albedo images, from multiple non-frontal images. Also a dense 3D shape model of the face is reconstructed, which can be used to generate faces under pose. In the authors’ experiments, the proposed method is able to improve the face recognition scores in more than 90% of the cases. Using the likelihood ratio framework, they show for the same experiment that for data initially unsuitable for forensic use, the reconstructions become meaningful in a forensic context in more than 60% of the cases.

Tangible 3D modeling of coherent and themed structures

We present CubeBuilder, a system for interactive, tangible 3D shape modeling. CubeBuilder allows the user to create a digital 3D model by placing physical, non-interlocking cubic blocks. These blocks may be placed in a completely arbitrary fashion and combined with other objects. In effect, this turns the task of 3D modeling into a playful activity that hardly requires any learning on the part of the user. The blocks are registered using a depth camera and entered into the cube graph where each block is a node and adjacent blocks are connected by edges. From the cube graph, we transform the initial cubes into coherent structures by generating smooth connection geometry for some edges of the graph. Based on an analysis of the cube graph, we identify subgraphs that match given graph templates. These subgraph templates map to predefined geometric refinements of the basic shape. This, in turn, allows the user to tangibly build structures of greater details than the blocks provide in and of themselves. We show a number of shapes that have been modeled by users and are indicative of the expressive power of the system. Furthermore, we demonstrate the scalability of the tangible interface which appears to be limited only by the number of blocks available.

A Novel Approach to Mugshot Based Arbitrary View Face Recognition

National Key Laboratory of Fundamental Science on Synthetic Vision,College of Computer Science, Sichuan University, Chengdu 610065, P. R. China(Received November 18, 2015 : revised February 11, 2016 : accepted March 4, 2016)Mugshot face images, routinely collected by police, usually contain both frontal and profile views. Existing automated face recognition methods exploited mugshot databases by enlarging the gallery with synthetic multi-view face images generated from the mugshot face images. This paper, instead, proposes to match the query arbitrary view face image directly to the enrolled frontal and profile face images. During matching, the 3D face shape model reconstructed from the mugshot face images is used to establish corresponding semantic parts between query and gallery face images, based on which comparison is done. The final recognition result is obtained by fusing the matching results with frontal and profile face images. Compared with previous methods, the proposed method better utilizes mugshot databases without using synthetic face images that may have artifacts. Its effectiveness has been demonstrated on the Color FERET and CMU PIE databases. Keywords : Mugshot-based face recognition, Arbitrary view face recognition, Three-dimensional face reconstructionOCIS codes : (100.5010) Pattern recognition; (100.0100) Image processing; (100.3008) Image recognition, algorithms and filters; (100.6890) Three-dimensional image processing