Wenxing Ye

University of Florida (UF) 2006 - Present

Major: Electrical and Computer Engineering

Graduate University of Chinese Academy of Science (GUCAS) 2004 - 2006

Enrolled in graduate program associated with the National Lab of Pattern Recognition, CAS

University of Science & Technology of China (USTC) 1999 - 2004

Major: Automation
Dual: Computer Science

Electrical & Computer Engineering, UF 2006 - Present

Image/Signal Processing

• Working on data reconstruction from its projections. Trying to improve the reconstruction accuracy through introducing different box splines (e.g., tensor product B-spline, 7-direction box spline on Cartesian lattice, linear and quintic box splines on body centered cubic (BCC) lattice) as the basis function and taking advantages of the property that box splines are closed under Radon transform.
• Working on generalizing the univariate sinc wavelets to image samples on arbitrary lattices based on the idea of Brillouin zone decomposition.
• Designed a new geometric framework for explicit derivation of multivariate sampling functions (sinc) on multidimensional lattices. This geometric approach also provides a frequency partition of the spectrum that leads to a non-separable extension of the 1-D Shannon (sinc) wavelets to the multivariate setting. Also, proposed a generalization of the Lanczos window function to provide a practical and unbiased approach for signal reconstruction on sampling lattices. This framework is general for lattices of any dimension and experiments on 3D body centered cubic (BCC) and face centered cubic (FCC) lattices validated the theoretical expectations about their superiority over the Cartesian lattice. This scheme was implemented as part of the vuVolume rendering package.
• Worked on developing the innovative cubic spline interpolation scheme on the BCC lattice based on its unique tetrahedralization. Designed the finite-differencing scheme on the BCC lattice with guaranteed order of accuracy. This cubic spline is comparable to the tensor-product Catmull-Rom spline on the Cartesian lattice since they are both C¹ interpolating and fourth-order methods. The interpolation algorithm is integrated to the vuVolume rendering package. Its implementation is simple and computation is efficient.
• Implemented a Fourier volume rendering algorithm with viable illumination model for MRI data visualization. Achieved interactive rendering for changing viewing point and light source.
• Worked on Poisson image editing which is actually image inpainting under guidance of a reference gradient field. A more computationally efficient generalized capacitance matrix algorithm was adopted and reduced the computational complexity from O(nlogn) to O(n).
• Investigated the feasibility of solid concentration measurement in liquid-solid two-phase flow using impedance spectroscopy. Built an online impedance analysis system with Agilent 4294A and gave preliminary results showing the relationship between the solid concentration and the conductance of the slurry at low and high frequencies.

Medical Imaging

• Improved the tomographic approach to reconstruction of diffusion propagators from diffusion weighted MRI data. Achieved higher reconstruction accuracy through introducing the interlaced multi-shell sampling geometry and body centered cubic (BCC) lattice which are more efficient in sampling the diffusion signal. In addition, proposed a novel box spline based tomographic reconstruction framework with higher approximation order. Experiments on mouse brain scans acquired from a 600MHz (14.1 Tesla) imaging spectrometer showed that the improved approach can recover more fiber crossings within areas of the deep gray and white layers of the superior colliculus.
• Experienced in real DW-MRI data analysis, including data formatting, denoising, motion correction through 3D image registration, fiber tracking and results visualization. Used Non-local means algorithm for denoising, different affine and non-rigid registration algorithms for motion correction, e.g. using cross cumulative residual entropy and using Gaussian mixture models on SIFT point sets.

Computer Vision

• Developed a target detection algorithm for Synthetic Aperture Radar (SAR) ground surveillance and tested on the public VHF change detection problem set released by US air force on SDMS. Applied change analysis, rotation invariant features and I-Relief feature weighting techniques to reduce the false alarm level. With the same number of correct detections, the proposed scheme gives only 11 false alarms compared with 86 given by the benchmark algorithm.
• Developed a room number recognition algorithm for an autonomous robot, Koolio. The number plate was located according to its characteristic of edge richness and the numbers was recognized through template matching.
• Implemented image segmentation demos based on generalized principle component analysis (GPCA) and active contour (snake) model with termination potentials and balloon factor.

Machine Learning

• Worked on large population speaker recognition problem. Developed a content/language independent demo system in C++ with CMU Sphinx toolkit. It models speech data with Gaussian Mixture Model (GMM) in Mel-Frequency Cepstral Coefficients (MFCC) feature space and does classification using Maximum Likelihood (ML) classifier. The demo system was tested on TIDIGITS dataset and by Narus Software on real recordings from hundreds of speakers and gave 100% recognition rate.
• Worked on text mining, specifically authorship identification problem using bag-of-words model, naive Bayes classifier and maximum margin feature weighting techniques. Lemur toolkit was used for feature extraction.
• Examined the possibility of detecting potential Parkinson patients through fractal analysis on real EEG (Electroencephalography) data collected from healthy persons and Parkinson patients. The Hurst parameter, phase space and Lyapunov exponent were analyzed and the Hurst parameter showed the best discrimination between healthy person and Parkinson patient.

National Lab of Pattern Recognition, Chinese Academy of Science 2004 - 2006

• Developed a real-time object detection system in C++ using high speed line-scan camera, specifically for the application of cotton impurity inspection. Focused on problems of color space discrimination and motion compensation.