Improved Classification of Blurred Images with Deep-Learning Networks Using Lucy-Richardson-Rosen Algorithm

Pattern recognition techniques form the heart of most, if not all, incoherent linear shift-invariant systems. When an object is recorded using a camera, information sampled by point spread function (PSF) system, replacing every with PSF in sensor. The sharp Kronecker Delta-like when numerical aperture (NA) large no aberrations. NA small, and system has aberrations, appears blurred. In case known, then blurred image can be deblurred scanning over intensity pattern looking for matching conditions through mathematical process called correlation. Deep learning-based classification computer vision applications gained attention recent years. probability highly dependent on quality images as even minor blur significantly alter results. this study, recently developed deblurring method, Lucy-Richardson-Rosen algorithm (LR2A), was implemented to computationally refocus presence spatio-spectral performance LR2A compared against parent techniques: Lucy-Richardson non-linear reconstruction. exhibited superior capability extreme cases Experimental results picture high-resolution smartphone cameras are presented. improve performances widely used deep convolutional neural networks classification.