Mingquan Zhou

3D face reconstruction from skull by regression modeling in shape parameter spaces

Abstract Craniofacial reconstruction is to estimate a person׳s face model from the skull. It can be applied in many fields such as forensic medicine, face animation. In this article, a regression modeling based method for craniofacial reconstruction is proposed, in which a statistical shape model is built for skulls and faces, respectively, and the relationship between them is extracted in the shape parameter spaces through partial least squares regression (PLSR). Craniofacial reconstruction is realized by using the relationship and the face statistical shape model. To better represent craniofacial shape variations and boost the reconstruction, both the skull and face are divided into five corresponding feature regions, and a mapping from each skull region to the corresponding face region is established. For an unknown skull, the five face regions are obtained through the five mappings, and the face is recovered by stitching the five face regions. The attributes such as age and body mass index (BMI) can be added into the mappings to achieve the face reconstruction with different attributes. Compared with other statistical learning based methods in literature, the proposed method more directly and reasonably reflects the relationship that the face shape is determined by the skull and influenced by some attributes. In addition, the proposed method does not need to locate landmarks, whose quantity and accuracy can highly affect the reconstruction. Experimental results validate the proposed method.

Segmentation algorithm of brain vessel image based on SEM statistical mixture model

Medical image segmentation has been a hot spot in recent years. And the segmentation of brain vessels image becomes a key-problem due to its complicated structure and small proportion. In this paper, the brain MRI images are processed with statistical analysis technology, and then the accuracy of segmentation is improved by the random assortment iteration .First the MIP algorithm is applied to decrease the quantity of mixing elements. Then the Gaussian Mixture Model is put forward to fit the stochastic distribution of the brain vessels and brain tissue. Finally, the SEM algorithm is adopted to estimate the parameters of Gaussian Mixture Model. The feasibility and validity of the model is verified by the experiment. With the model, small branches of the brain vessel can be segmented, the speed of the convergent is improved and local minima are avoided.

Calibrating effective focal length for central catadioptric cameras using one space line

In camera calibration, focal length is the most important parameter to be estimated, while other parameters can be obtained by prior information about scene or system configuration. In this paper, we present a polynomial constraint on the effective focal length with the condition that all the other parameters are known. The polynomial degree is 4 for paracatadioptric cameras and 16 for other catadioptric cameras. However, if the skew is 0 or the ratio between the skew and effective focal length is known, the constraint becomes a linear one or a polynomial one with degree 4 on the effective focal length square for paracatadioptric cameras and other catadioptric cameras, respectively. Based on this constraint, we propose a simple method for estimation of the effective focal length of central catadioptric cameras. Unlike many approaches using lines in literature, the proposed method needs no conic fitting of line images, which is error-prone and highly affects the calibration accuracy. It is easy to implement, and only a single view of one space line is enough with no other space information needed. Experiments on simulated and real data show this method is robust and effective.

3D craniofacial reconstruction using reference skull-face database

Craniofacial reconstruction is aiming at estimating the outlook of an unknown or an unidentified skull. In this paper, we present an approach of the craniofacial reconstruction of Chinese people. Firstly, we build a skull-face database and classify it in terms of age, area and gender. Then Thin-Plate spline (TPS) is adopted to achieve non-rigid registration between the unidentified skull and reference skull. however, the craniofacial result is determined by the choice of reference template. Here a statistic method is adopted to estimate outlook from subclass of skull-face database using Principle component analysis. In order to improve the accuracy of the result, we select the suitable organ (eyes, nose and mouth) for the statistic result based on anatomy principle from the database and achieve the organ and face integration to build the final outlook. Finally, we show some experiments of the algorithm.

Parallel techniques for improving three-dimensional models storing and accessing performance

Nowadays, the volume of multimedia and unstructured data has grown rapidly. More and more three-dimensional (3D) models are created for ever increasing applications. New storage and processing technologies are needed to keep pace with the continuous growth of big data. Hadoop is an attractive and open-source platform for large-scale data storage and analytics. Our previous research work has applied Hadoop distributed file system to efficiently manage 3D data for a 3D model retrieval system. To take better advantages of Hadoop, in this paper we propose two parallel strategies to improve the storing and accessing performance of 3D models. The MapReduce paradigm is adopted to provide a coarse grained parallelism for data loading, and a lightweight multithreaded algorithm is presented for data accesses. We conduct an extensive performance study on a cluster and the results show that significant performance increase can be gained for the parallel techniques.

Automated spectral classification using template matching

An automated spectral classification technique for large sky surveys is proposed. We firstly perform spectral line matching to determine redshift candidates for an observed spectrum, and then estimate the spectral class by measuring the similarity between the observed spectrum and the shifted templates for each redshift candidate. As a byproduct of this approach, the spectral redshift can also be obtained with high accuracy. Compared with some approaches based on computerized learning methods in the literature, the proposed approach needs no training, which is time-consuming and sensitive to selection of the training set. Both simulated data and observed spectra are used to test the approach; the results show that the proposed method is efficient, and it can achieve a correct classification rate as high as 92.9%, 97.9% and 98.8% for stars, galaxies and quasars, respectively.

A rendering algorithm based on ray-casting for medical images

Ray-casting algorithm is usually used in medical volume visualization, but there are a lot of limitations inherent to this algorithm, like high computational demand, low quality images, or a fixed classification. In contrast, real-time high quality volume rendering algorithms are still a challenge nowadays. We introduce here an efficient and accurate volume rendering algorithm for medical images, and this algorithm can achieve up to 2.5 fps on PCs for 512 times 512 times 482 times 2 Byte volume. The algorithm classifies volume data into the foreground voxels and background voxels, and the former voxels are resampled using the similar LOD (Layer of Detail) technique, then optical attribute of the resampling points is determined by a new method. This method is reasonably associated with the distance between the object and the viewpoint, as well as the object and the light source. In addition, the background voxels are displayed by the accelerated method based on space leaping resampling. The results show that interactive volume rendering can be implemented for most medical volume data on PCs. Meanwhile, tissues or organs can be displayed clearly, which is more coincident with the human vision.

Closed cube computation on multi-core CPUs

Closed cube computation is a popular method to solve the huge output problem of data cubing and has attracted great interests among researchers. A lot of efficient algorithms such as QC-DFS, C-Cubing and CC ALG have been proposed. However, due to significant changes in computer hardware architecture in recent years, these algorithms need to be revisited to achieve good performance. In this paper, we present a parallel algorithm for closed cube computation on multi-core CPUs to fully utilize the processor resources. In our algorithm, a cost-based load balance strategy is designed to deal with data skews and partitions based on each individual dimension are simultaneously computed to generate closed cells. An extensive performance study is conducted and the results show that good speedups can be gained for the parallel algorithm.

A new easy calibration algorithm for para-catadioptric cameras

In this paper, a new easy calibration method for para-catadioptric cameras is proposed. We derive a nonlinear constraint on all camera intrinsic parameters from the projections of any three collinear space points on the viewing sphere. With the principal point known, the constraint becomes linear on all other intrinsic parameters which are the effective focal length, the aspect ratio and the skew, so the three parameters can be estimated linearly by using SVD. The principal point can be well determined from the bounding ellipse of the catadioptric image in real applications. Unlike approaches using lines in literature, the proposed method needs no conic fitting, which is hard to accomplish and highly affects the accuracy of the calibration. Experiments on simulated and real data show this method is robust and effective.

Novel cross correlation method for redshift determination of galaxy spectra

The rapid development of the astronomical observation has led to many large sky surveys such as SDSS, 2DF, LAMOST etc. Because of the sheer size of these surveys, it becomes urgent to develop methods of reliable and automated spectral recognition. A new cross correlation technique for redshift determination of galaxy spectra is presented in this paper. We use principle components analysis to construct galaxy templates. According to the redshift candidates determined by spectra line features, cross-correlation between the observed spectrum and the templates is measured by the weighted sum of several similarity evidences. The candidate of the highest correlation is chosen as the estimated redshift. Both simulated spectra and observed spectra are used to test the proposed method, the correct rate can reach 97% above.