Houston, Texas, United States
Developing novel imaging and image processing solutions for Samsung mobile platforms
My research focuses on designing algorithms and hardware for novel compressive hyperspectral imaging systems for microscopy and for 3D hyperspectral reconstruction - Designed and implemented a novel training algorithm for multi-rate neural networks for compressed domain image classification on compressive samples acquired by the single-pixel camera. The multi-rate network is capable of predicting over a range of compressive measurement rates instead of a single measurement rate. - Designed and built a compressive hyperspectral reflection/transmission microscope using a single spectrometer as detector and used it to measure spectra of gold nanoparticles and fluorescent microspheres. The system is low-cost and fast compared to conventional hyperspectral microscopy methods. - Designed and built a low-cost hyperspectral projector system that generates programmable hyperspectral structured illumination based on a single digital micromirror device. The system is used for reconstruction of hyperspectral 3D shape and reconstruction of 3D volume density of participating media. - Developed a deep learning tool for classifying digital images of brain tissue slides into benign or malicious tumor tissues. The dataset is from an online library of brain tumor patients.
Worked on image and video subjective quality assessment through deep learning and digital signal processing: - Designed a deep learning tool for subjective assessment of video quality with raw 1080p frames as input - Implemented the model training and inference pipeline with Torch 7 - Achieved state-of-the-art performance compared to popular objective video quality metrics - Collected and synthesized a large image dataset with automatic labelling - Developed a deep learning tool for predicting downscaling factor of images with added noise on the synthesized image dataset - Implemented the model training and inference pipeline with Torch 7 - Achieved closed to 100% inference accuracy
Research on particle physics - collaborated with top researcher as part of the global effort in the search for supersymmetry particles - Analyzed large amount of data from LHC using C++