Ptychography is a state-of-the-art imaging technique that enables high-resolution and quantitative phase imaging by reconstructing complex-valued images from diffraction patterns. This project successfully developed a novel multiresolution framework to address the computational challenges of ptychographic imaging, including non-convex optimization problems and high-dimensional data processing. The framework employs a hierarchical approach using box-splines, allowing images to be refined progressively from coarse to fine scales. This method dramatically improved computational efficiency, accelerated convergence, and reduced the risk of local minima during optimization. Hessian Total Variation (HTV) regularization was integrated into the framework to ensure stability and promote sparsity in second-order derivatives, enabling precise and reliable reconstructions. The framework was applied to both linear and nonlinear problems, showcasing its adaptability and robustness in solving complex phase retrieval tasks. Key innovations included optimizing probe parameters and overlap ratios, which enhanced the recovery of phase information and minimized artifacts in the reconstructed images. Extensive experimental validation demonstrated the effectiveness of the approach, with significant improvements in computational performance and reconstruction quality compared to traditional methods. The project achieved superior convergence performance, especially in challenging nonlinear scenarios. This multiresolution framework establishes a powerful and efficient solution for high-resolution ptychographic imaging, advancing capabilities in fields such as biological imaging and materials science.
- Jonathan Dong, [email protected], EPFL
- Vincent Guillemet, [email protected], EPFL