Iman Yi Liao

A fractal-based relaxation algorithm for shape from terrain image

We consider the problem of extracting surface shape from a single terrain image. Although fractal models play an important role in simulating terrain models, the various Shape-from-Shading (SFS) techniques that have been applied to this kind of problem have not been coupled with a fractal prior. In this paper, we define the SFS problem of terrain imaging as a fractal-regularized problem, and solve it using Maximum-A-Posterior (MAP) estimation. In addition, we also propose a relaxation algorithm based on Landweber iteration in order to solve it. The optimum terrain surface corresponding to the observed image does not have to be the convergent result. The result can be picked up during the process of iteration with the number of iterations specified by an image-based estimation method proposed in this paper. Experimental results on both simulated data and real data show that our algorithm can efficiently extract terrain surfaces, and is more accurate than some well-known SFS algorithms, including the Horn, Zheng-Chellappa, Tsai-Shah, Pentland linear, and Lee-Rosenfeld methods.

Effect of facial feature points selection on 3d face shape reconstruction using regularization

This paper aims to test the regularized 3D face shape reconstruction algorithm to find out how the feature points selection affect the accuracy of the 3D face reconstruction based on the PCA-model. A case study on USF Human ID 3D database has been used to study these effect. We found that, if the test face is from the training set, then any set of any number greater than or equal to the number of training faces can reconstruct exact 3D face. If the test face does not belong to the training set, it will hardly reconstruct the exact 3D face using 3D PCA-based models. However, it could reconstruct an approximate face shape depending on the number of feature points and the weighting factor. Furthermore, the accuracy of reconstruction by a large number of feature points (> 150) is relatively the same in all cases even with different locations of points on the face. The regularized algorithm has also been tested to reconstruct 3D face shapes from a number of feature points selected manually from real 2D face images. Some 2D images from CMU-PIE database have been used to visualize the resulted 3D face shapes.

A neural network-based point registration method for 3D rigid face image

Intelligent detection of human face image combined with the real-time video monitoring has been applied to improve the secure and protective possibility. The registration is an indispensible step before distinguishing the variation among the images. Neural network (NN) has a strong learning ability from a mass unstructured point cloud even containing noisy data. Neural network has been applied to register 3D reconstructed ear data and 3D surface of bunny and to achieve the better results. Motivated by this idea, NN-based registration method for 3D rigid face image is proposed. This paper presented the proof process of obtaining rotation matrix and translation vector according to the training process of neural network. Then the measure index of registration performance was provided. The elaborate experiments were conducted on the 3D USF face database (provided by the University of South Florida) to verify the effectiveness of neural network as a registration method. Next, two comparisons were performed, one group was NN-based and ICP-based registration methods and the other group was our proposed NN-based and other NN-based registration methods. The experimental results showed that neural network is a robust and potential registration method for rigid face image registration. Furthermore, our proposed NN-based registration method is extended easily to do 2D-to-3D registration and non-rigid face registration.

Adaptive face modelling for reconstructing 3D face shapes from single 2D images

Example-based statistical face models using principle component analysis (PCA) have been widely deployed for three-dimensional (3D) face reconstruction and face recognition. The two common factors that are generally concerned with such models are the size of the training dataset and the selection of different examples in the training set. The representational power (RP) of an example-based model is its capability to depict a new 3D face for a given 2D face image. The RP of the model can be increased by correspondingly increasing the number of training samples. In this contribution, a novel approach is proposed to increase the RP of the 3D face reconstruction model by deforming a set of examples in the training dataset. A PCA-based 3D face model is adapted for each new near frontal input face image to reconstruct the 3D face shape. Further an extended Tikhonov regularisation method has been employed to reconstruct 3D face shapes from a set of facial points. The results justify that the proposed adaptive PCA-based model considerably improves the RP of the standard PCA-based model and outperforms it with a 95% confidence level.

Can Computer Vision Techniques be Applied to Automated Forensic Examinations? A Study on Sex Identification from Human Skulls Using Head CT Scans

Sex determination from human skeletal remains is a challenging problem in forensic anthropology. The human skull has been regarded as the second best predictor of sex because it contains several sexually dimorphic traits. Previous studies have shown that morphological assessment and morphometric analysis can be used to assess sex variation from dried skulls. With the availability of CT scanners, the field has seen increasing computer aided techniques in assisting these traditional forensic examinations. However, they largely remain at the level of providing a digital interface for landmarking for morphometric analysis. A recent research has applied shape analysis techniques for morphological analysis on a specific part of the skull. In this paper, we endeavor to explore the application of computer vision techniques that have prominently been used in the field of 3D object recognition and retrieval, for providing alternative method to achieve sex identification from human skulls automatically. We suggest a possible framework for the whole process including multi-region representation of the skull with 3D shape descriptors, and particularly examined the role of 3D descriptors on sex identification accuracy. The experimental results on 100 head post mortem CT scans indicate the potential of 3D descriptors for skull sex classification. To the best of our knowledge, this is the first work to have approached skull sex prediction from this novel perspective.

Reconstructing 3D Face Shapes from Single 2D Images Using an Adaptive Deformation Model

The Representational Power (RP) of an example-based model is its capability to depict a new 3D face for a given 2D face image. In this contribution, a novel approach is proposed to increase the RP of the 3D reconstruction PCA-based model by deforming a set of examples in the training dataset. By adding these deformed samples together with the original training samples we gain more RP. A 3D PCA-based model is adapted for each new input face image by deforming 3D faces in the training data set. This adapted model is used to reconstruct the 3D face shape for the given input 2D near frontal face image. Our experimental results justify that the proposed adaptive model considerably improves the RP of the conventional PCA-based model.

Finite difference error analysis of geometry properties of implicit surfaces

In this paper we investigate the impact of different finite difference method (FDM) on estimation of the curvatures and other geometry properties of an implicit surface extracted from volumetric data. The FDM estimation case studies of the geometrical features are conducted on the surface in different resolutions which are refined surface, medium surface and coarse surface. A ground truth is established on each resolution to be compared with the experiment results. Then we apply relative error analysis on the results to find the influences of FDM that can be utilized in the estimation.

Skull Registration Using Rigid Super-Curves

This research presents an algorithm, Rigid Super Curves (RSC), to solve the problem of registering two sets of digitized skulls data under a rigid transformation using crest lines. The method that restitutes the rigid transformation between two sets of fully matched curves is propounded. RSC exploits the non-ambiguity of B-Spline representation of super-curves whilst overcoming the inability of super-curves to restore rigid transformations. A further contribution of this study is a two-stage algorithm based on RSC which registers two sets of partially matched curves under a rigid transformation. The algorithm improves the robustness over feature based methods by considering the structure rather than individual points of the curve. Experimental results on CT scanned skull data show that proposed algorithm is more robust and accurate at restoring the rigid transformation between two sets of crest line data compared with Iterated Closest Point and Super Curves methods.

Isosurface Extraction of Volumetric Data Using Implicit Surface Polygonization

Isosurface extraction is always an interesting topic in the field of computer graphics and scientific visualization. This study surveys isosurface construction methods, particularly in implicit surface polygonization using tracking partition. An investigation is conducted on the implementation of it. With reference of original Jules Bloomenthal’s implementation, an extension of the implicit surface polygonization is developed with concern of real volumetric data. Rather than conventional methods, such as Marching Cubes and its variants, the method has advantages on time complexity, flexibility and is easy to implement. A brief comparison has been done among these methods. During the development, some practical issues, such as starting point placement, seed point generation, manifold surface, manifold-with-boundary, are addressed. Possible solutions of these problems are suggested. Some ideas on enhancement of the method are conveyed at end of this paper for further work.

Bi-Stage Large Point Set Registration Using Gaussian Mixture Models

Point set registration is to determine correspondences between two different point sets, then recover the spatial transformation between them. Many current methods, become extremely slow as the cardinality of the point set increases; making them impractical for large point sets. In this paper, we propose a bi-stage method called bi-GMM-TPS, based on Gaussian Mixture Models and Thin-Plate Splines (GMM-TPS). The first stage deals with global deformation. The two point sets are grouped into clusters independently using K-means clustering. The cluster centers of the two sets are then registered using a GMM based method. The point sets are subsequently aligned based on the transformation obtained from cluster center registration. At the second stage, the GMM based registration method is again applied, to fine tune the alignment between the two clusters to address local deformation. Experiments were conducted on eight publicly available datasets, including large point clouds. Comparative experimental results demonstrate that the proposed method, is much faster than state-of-the-art methods GMM-TPS and QPCCP (Quadratic Programming based Cluster Correspondence Projection); especially on large non-rigid point sets, such as the swiss roll, bunny and USF face datasets, and challenging datasets with topological ambiguity such as the banana dataset. Although the Coherent Point Drift (CPD) method has comparable computational speed, it is less robust than bi-GMM-TPS. Especially for large point sets, under conditions where the number of clusters is not extreme, a complexity analysis shows that bi-GMM-TPS is more efficient than GMM-TPS.