Yongjian Xi

Thalamus segmentation from diffusion tensor magnetic resonance imaging

In this paper, we propose a semi-automatic thalamus and thalamus nuclei segmentation algorithm from diffusion tensor magnetic resonance imaging (DT-MRI) based on the mean-shift algorithm. Comparing with existing thalamus segmentation algorithms which are mainly based on K-means algorithm, our mean-shift based algorithm is more flexible and adaptive. It does not assume a Gaussian distribution or a fixed number of clusters. Furthermore, the single parameter in the mean-shift based algorithm supports hierarchical clustering naturally

An Automation System of Rooftop Detection and 3D Building Modeling from Aerial Images

This paper presents a prototype system of rooftop detection and 3D building modeling from aerial images. In this system, without the knowledge of the position and orientation information of the aerial vehicle a priori, the parameters of the camera pose and ground plane are first estimated by simple human–computer interaction. Next, after an over-segmentation of the aerial image by the Mean-Shift algorithm, the rooftop regions are coarsely detected by integrating multi-scale SIFT-like feature vectors with SVM-based visual object recognition. 2D cues alone however might not always be sufficient to separate regions such as parking lots from building roofs. Thus in order to further refine the accuracy of the roof-detection result and remove the misclassified non-rooftop regions such as parking lots, we further resort to 3D depth information estimated based on multi-view geometry. More specifically, we determine whether a candidate region is a rooftop or not according to its height information relative to the ground plane, whereas the candidate region’s height information is obtained by a novel, hierarchical, asymmetry correlation-based corner matching scheme. The output of the system will be a water-tight triangle mesh based 3D building model texture mapped with the aerial images. We developed an interactive 3D viewer based on OpenGL and C+ + to allow the user to virtually navigate the reconstructed 3D scene with mouse and keyboard. Experimental results are shown on real aerial scenes.

Integrated modeling of PET and DTI information based on conformal brain mapping

Recent advances in imaging technologies, such as Magnetic Resonance Imaging (MRI), Positron Emission Tomography (PET) and Diffusion Tensor Imaging (DTI) have accelerated brain research in many aspects. In order to better understand the synergy of the many processes involved in normal brain function, integrated modeling and analysis of MRI, PET, and DTI is highly desirable. Unfortunately, the current state-of-art computational tools fall short in offering a comprehensive computational framework that is accurate and mathematically rigorous. In this paper we present a framework which is based on conformal parameterization of a brain from high-resolution structural MRI data to a canonical spherical domain. This model allows natural integration of information from co-registered PET as well as DTI data and lays the foundation for a quantitative analysis of the relationship between diverse data sets. Consequently, the system can be designed to provide a software environment able to facilitate statistical detection of abnormal functional brain patterns in patients with a large number of neurological disorders.

Virtual navigation of interior structures by lidar

In this project, we propose to develop a prototype system that can automatically reconstruct 3D scenes of the interior of a building, cave or other structure using ground-based LIDAR scanning technology. We develop a user-friendly real-time visualization software package that will allow the users to interactively visualize, navigate and walk through the room from different view angles, zoom in and out, etc.

A Region-Growing Based Iso-Surface Extraction Algorithm

In this paper, we propose a new region-growing based iso-surface extraction algorithm that can generate high-quality curvature-adaptive semi-regular meshes, preserve sharp features and will extract all the disjoint components of the iso-surface. More importantly, in this paper, we propose a novel normal consistency constraint that ensures the intersection of the Delaunay sphere of the new triangle and the iso-surface is a topological disk, an important property that makes the new algorithm very robust when dealing with large scale of volumetric datasets of complex topology and geometry.

Thalamus segmentation from diffusion tensor magnetic resonance imaging.

In this paper, we propose a semi-automatic thalamus and thalamus nuclei segmentation algorithm from diffusion tensor magnetic resonance imaging (DT-MRI) based on the mean-shift algorithm. Comparing with existing thalamus segmentation algorithms which are mainly based on K-means algorithm, our mean-shift based algorithm is more flexible and adaptive. It does not assume a Gaussian distribution or a fixed number of clusters. Furthermore, the single parameter in the mean-shift based algorithm supports hierarchical clustering naturally.

Brain structure segmentation from MRI by geometric surface flow.

We present a method for semiautomatic segmentation of brain structures such as thalamus from MRI images based on the concept of geometric surface flow. Given an MRI image, the user can interactively initialize a seed model within region of interest. The model will then start to evolve by incorporating both boundary and region information following the principle of variational analysis. The deformation will stop when an equilibrium state is achieved. To overcome the low contrast of the original image data, a nonparametric kernel-based method is applied to simultaneously update the interior probability distribution during the model evolution. Our experiments on both 2D and 3D image data demonstrate that the new method is robust to image noise and inhomogeneity and will not leak from spurious edge gaps.

An iterative surface evolution algorithm for multiview stereo

We propose a new iterative surface evolution algorithm for multiview stereo. Starting from an embedding space such as the visual hull, we will first conduct robust 3D depth estimation (represented as 3D points) based on image correlation. A fast implicit distance function-based region growing method is then employed to extract an initial shape estimation based on these 3D points. Next, an explicit surface evolution will be conducted to recover the finer geometry details of the recovered shape. The recovered shape will be further improved by several iterations between depth estimation and shape reconstruction, similar to the Expectation Maximization (EM) approach. The experiments on the benchmark datasets show that our algorithm can obtain high-quality reconstruction results that are comparable with the state-of-art methods, with considerable less computational time and complexity.

Rooftop Detection and 3D Building Modeling from Aerial Images

This paper presents a new procedure for rooftop detection and 3D building modeling from aerial images. After an over-segmentation of the aerial image, the rooftop regions are coarsely detected by employing multi-scale SIFT-like features and visual object recognition. In order to refine the detected result and remove the non-rooftop regions, we further resort to explore the 3D information of the rooftop by 3D reconstruction. Wherein, we employ a hierarchical strategy to obtain the corner correspondence between images based on an asymmetry correlation corner matching. We determine whether a candidate region is a rooftop or not according to its height information relative to the ground plane. Finally, the 3D building model with texture mapping based on one of the images is given. Experimental results are shown on real aerial scenes.

A nonparametric approach for noisy point data preprocessing

3D point data acquired from laser scan or stereo vision can be quite noisy. A preprocessing step is often needed before a surface reconstruction algorithm can be applied. In this paper, we propose a nonparametric approach for noisy point data preprocessing. In particular, we proposed an anisotropic kernel based nonparametric density estimation method for outlier removal, and a hill-climbing line search approach for projecting data points onto the real surface boundary. Our approach is simple, robust and efficient. We demonstrate our method on both real and synthetic point datasets.