Convolutional Neural Networks to Decode Images in a Wavefront Coding Imaging System

Abstract

Wavefront coding technique has been used to extend the depth of focus in an optical imaging system. An optical element called a phase mask allows coded images to be obtained since the point spread function remains almost invariant in an axial range. Subsequently, a computational technique is required to decode the acquired images. An optical-computational technique is proposed to use a phase mask for the coding stage and a convolutional neural network for the final restoration. Comparative results are made between cubic and trefoil profile phase masks and decoded using the traditional Wiener filter and a convolutional neural network. Image quality evaluation is done using the peak signal-to-noise ratio.