Information retrieval from optical speckles is desired yet challenging. Insufficient sampling, especially in sub-Nyquist domain, of speckles significantly destroys the encoded information and correlations among these speckle grains. To address that, we trained a deep neural network to combat the physical imperfection: the sub-Nyquist sampled speckles (~14 below the Nyquist criterion) are interpolated up to a well-resolved level (322 times more pixels to resolve the same FOV) with smoothed morphology fine-textured, and more importantly, lost information retraced. With the FOV-resolution dilemma favorably overcome, it deepens our understanding of the scattering, enabling big and clear imaging in complex scenarios.
|