Light Field Image Quality Assessment with Dense Atrous Convolutions

Unlike regular images that represent only light intensities, Light Field (LF) contents carry information about the intensity of light in a scene, including the direction in which light rays are traveling in space. This allows for a richer representation of our world, but requires large amounts of data that need to be processed and compressed before being transmitted to the viewer. Since these techniques may introduce distortions, the design of Light Field Image Quality Assessment (LF-IQA) methods is important. Currently, most LF-IQA methods use traditional 2D image quality assessment techniques or rely on low-level spatial features. In this paper, we propose a novel no-reference LF-IQA method that takes into account both LF angular and spatial information. The proposed method is made up of two processing streams with identical blocks of Convolutional Neural Network (CNN), Atrous Convolution layers (ACL), and a regression block for quality prediction. The results show that the method is robust and outperforms current state-of-the-art methods.

1. Introduction

A Light Field Image (LFI) is a four-dimensional (4D) signal representation that is defined by a plenoptic 4D function L(u,v,s,t), where the coordinates (u,v) and (s,t) correspond to the angular domain and the spatial domain, respectively [1]. LFIs are rendered and displayed using different formats, such as subaperture images (SAIs) or horizontal/vertical epipolar plane images (EPIs). SAIs are two-dimensional (2D) slices acquired by gathering samples at fixed (s,t) coordinates, carrying pixels from a specific view that reflect spatial information of the LFI. Vertical EPIs can be obtained by fixing the coordinates u and s, while horizontal EPIs can be obtained by fixing the coordinates v and t. EPI describes the shift of the pixel information over an angular axis, including both spatial and angular dimensions. In case of discontinuity in the angular domain, the lines on the EPI become distorted.