Haichuan Zhang
Ongoing Project
Deep, Convergent, Unrolled Non-Blind Image Deconvolution
Information Processing and Algorithms Lab - Pennsylvania State University, University Park
We propose a deep, interpretable neural network by unrolling the widely-used Half-Quadratic Splitting (HQS) algorithm. A structured parametrization scheme is introduced to ensure convergence with minimal impact on network performance.
The convergence of this neural network, under our parametrization, is both theoretically established and empirically validated through simulations.
In comparison with SOTA, our approach outperforms both traditional iterative algorithms and contemporary deep neural networks by approximately 1 dB in PSNR and 0.1 in SSIM, all while ensuring convergence and maintaining interpretability.
The paper titled ”A Convergent Neural Network for Non-Blind Image Deblurring” has been accepted at the 2023 IEEE International Conference on Image Processing (ICIP). Additionally, ”Deep, Convergent, Unrolled Non-Blind Image Deconvolution” has been submitted to the Transactions on Computational Imaging.
High-Resolution Transcranial Ultrasound Neuromodulation at Large Scale
Information Processing and Algorithms Lab - Pennsylvania State University, University Park
Designed a neural network model to analyze ultrasound scanning images.
Engineered the network to determine the time delays for each emitter in the Ultrasound Phased Arrays for Large-Scale Ultrasound Neuromodulation.
• The purpose of the determined time delays was to ensure maximum pressure at desired locations within non-uniform media, optimizing neuromodulation.
Scene Segmentation-Guided Lens Mapping for Bokeh Effect Transformation
Information Processing and Algorithms Lab - Pennsylvania State University, University Park
Developed the Segmentation-Guided Lens Mapping for Bokeh Effect Transformation (SGLM) methodology, which comprises the Foreground Segmentation Module (FSM) and the Lens Mapping Module (LMM), to emphasize the unique optical properties of different lenses.
Designed the FSM to accurately predict the foreground alpha matte using its Semantic Prediction Branch and Detail Prediction Branch. This ensures sharpness preservation in the foreground while transforming the bokeh effect in the out-of-focus region.
Implemented the LMM with multiple encoders and decoders, allowing for the transformation of bokeh effects between various lenses. Each encoder and decoder was tailored for a specific lens.