We are now opening a Post-Doctoral position at UoS-NPU Joint Lab. You will be under joint supervisionof Prof. J.W.McBride at University of Southampton, UK and Prof. Jingdong Chen at Northwester Polytechnic University, China. If you are interested, please send: 1) your CV, and 2) one-page personal statement to y.yang@nwpu.edu.cn by 17th April, 2020.
Location:1 year at NWPU, and 2 years at UoS.
Project title:Data Registration for the evaluation of low level (sub-visual) wear on biological surfaces.
Skills required:The successful candidate will have a combination of good communication skills and aPh.D in Engineering or Physics. The project will require exceptional experimental and analytical skills,as evidenced by existing journal publications. Candidates with particular skills in image processing andor signal processing would be encouraged to apply. The software skills required are MATLAB, and C++or C#. The candidate can be expected to work in a development environment with other teammembers.
Despite rapid advances in precision surface engineering, there are a number of areas wherecharacterisation is problematic. These include free-form, structured and aspheric surfaces. Earlydetection of wear of biological surfaces is critical for the prevention of degradation in many areas ofhealth care and beyond. The proposed research addresses the detection and characterisation of earlywear on free-form surfaces. Teeth are complex free-form surfaces which wear at non-uniform rates.
The results of this study will be applicable to a wide range of applications which require measurementof complex 3D surfaces over time. To solve this challenging problem, the following underlying sciencemust be addressed:
1. Measurement. Complex 3D surfaces must be measured to a high data density, withdata/voxel samples <5 µm, over volume cubes of the order of 10 mm. On the defined scale,the individual 3D data should be to a sub-micron resolution and uncertainty.
2. Registration. The 3D data should be registered without any defining features, such that twosurfaces, measured before and after wear or change can be subtracted, to a defined uncertainty, so change can be quantified.
This research project will focus on the Registration Problem. Registration consists of aligning pointclouds using specifific sets of transformations. This problem appears in many difffferent settings such asoptical character recognition, augmented reality and aligning data from magnetic resonance imagingwith computer aided tomography scans and within the physical sciences investigating change on alarge scale. Most often, the registration problem is divided into two difffferent classes: rigid and nonrigid registration. The rigid registration is concerned with fifinding affiffiffine transformations that preservedistances and the non-rigid registration problem includes fifinding general as well as non-lineartransformations, where the body can change shape, requiring shearing and scaling of data. Severalapproaches have been devised for the various types of registration problems that appear in image andpattern analysis. The main approaches are as follows. The Iterative Closest Point (ICP) algorithm [1], isone of the most widely used methods, iteratively revising the transformation (translation & rotation)to minimize the sum of squared differences between the coordinates of the matched pairs. As such,it works best if the fifirst point cloud is suffiffifficiently close to the second. Robust Point Matching (RPM),[2], followed based on the soft correspondence between data points. There are many otherapproaches used, some of which will be benchmarked on real worn surface data in the researchproposed. These include, Thin Plate Spline Robust Point Matching (TPS-RPM), and a Kernel Correlation(KC) approach, [3]. A Gaussian Mixture Model has been proposed in place of KC, as this has beenshown to be more effiffifficient. Coherent Point Drift (CPD) takes a probabilistic approach to aligning pointsets, similar to the Gaussian based registration method. In a recent study linked to surface wear, [4],a selective alignment approach was used based on the automated identification of referencelandmarks with a Scale Invariant Feature Transform (SIFT) used to identify the landmarks. It was shownto offer only limited success, further emphasising the need for this study.The data gathered in this study will not have exact point matching and there will be systematic errors(measurement uncertainty), random noise (surface roughness), and low-level surface wear (change ofform). The proposed solution developed in the feasibility study is to develop a manifold estimationand matching method, currently only tested on simulated data sets. The method fits a 3D polynomial(manifold) which is used to generate a curvature map, and then uses this map to compute thetransformation; the estimator of the manifold is noise tolerant using a median of means technique.
The new method is referred to as the Curvature and real algebraic manifold matching (CRAMM). In anew development Bezier surface fitting will also be used.
1. Develop code to implement the CRAMM method, and then compare the polynomial fittingapproach with the newly developed Bezier 3D surface method. Deliverable 1 joint journal paper.
2. Apply the methods developed to free form surfaces with roughness, to determine the repeatabilityof the process. Data will be provided from a XYRIS 2020 optical scanner. This will be extended to testreproducibility with the sample removed from the optical scanner and then replaced withoutrelocation to the same position.
3. Use the optimised code will be used to undertake a study related to sub-visual wear on dentalfreeform surfaces. A wear process will be instigated under controlled conditions, and samplesmeasured both optically and with X-Ray methods.
Note. This project is linked to a large research grant application (£1.3 Million) submitted in August2019. If awarded the researcher would be working with 2 other research assistants in UK, one at UoSand one at Kings College London.
英才招聘