Illumination pattern optimization in Fourier ptychographic microscopy using LED preselection strategy and simulated annealing algorithm

Abstract Fourier ptychographic microscopy (FPM) is a computational imaging technique that obtains images with high-resolution and wide field-of-view images. However, redundant image acquisition and the time-consuming reconstruction process limit both efficiency and image quality in practical application. To address these challenges, we propose an optimized illumination scheme that integrates spectral contribution analysis with simulated annealing (SA) algorithm. This method aims to reduce redundancy in the spectrum—such as contributions dominated by dark-field noise—while maintaining reconstruction quality and improving computational efficiency. Its effectiveness is evaluated through both simulations and experimental validations. Specifically, we first perform spectral contribution-based LED preselection by evaluating the frequency-domain contribution of each sub-image, retaining only those LED positions that positively impact the reconstructed image to form the initial illumination matrix. Building on this, we apply a SA algorithm to iteratively optimize the LED illumination pattern. This stochastic search technique explores the solution space through probabilistic perturbations to approximate a globally optimal configuration. Furthermore, the proposed iterative strategy exhibits adaptability, enabling optimal sampling schemes for different sample types. Experimental results demonstrate that the proposed method significantly accelerates the FPM reconstruction process while preserving high spatial resolution. Compared with the classical EPRY-FPM scheme, it reduces image reconstruction time by 50.9% with approximate reconstruction quality. Based on different illumination patterns and phase reconstruction methods, more comparative experiments and discussions are adopted, indicating that the proposed method not only could efficiently achieve high reconstruction quality, but also could be treated as a accelerated strategy for different phase reconstruction methods.