Image Reconstruction for Lensless Imaging Using a Phase Congruency Perception Model

Lensless imaging systems capture images without the need for traditional lenses. Instead, they employ flat and thin micro-optical devices, such as masks, diffusers, Fresnel zone plates, and other diffractive optical elements, to capture information-encoded images of imaging targets, and rely on advanced computational methods to reconstruct regular images from the encoded ones. Typical image reconstructions in lensless imaging utilize compressive sensing and sophisticated iterative optimization algorithms. Recently such iterative methods have been replaced by deep neural networks. However, in both approaches the cost (or loss) functions usually are defined on certain simple mathematical distance metric, without considering the inherent perceptive properties of images. In this report, a new method is proposed to construct cost functions for image reconstruction in lensless imaging. It is based on a visual perception model of image phase congruency, a measure of the consistency of the phase information across different spatial orientations and scales in images. The proposed method is demonstrated to provide improved image reconstruction quality using more than 200 encoded images from a virtual lensless imaging system with a Fresnel zone plate, as well a small number of encoded images from a physical one with an optical diffuser.