Frontiers in Signal Processing (Aug 2022)

Blind visual quality assessment of light field images based on distortion maps

  • Sana Alamgeer,
  • Mylène C. Q. Farias

DOI
https://doi.org/10.3389/frsip.2022.815058
Journal volume & issue
Vol. 2

Abstract

Read online

Light Field (LF) cameras capture spatial and angular information of a scene, generating a high-dimensional data that brings several challenges to compression, transmission, and reconstruction algorithms. One research area that has been attracting a lot of attention is the design of Light Field images quality assessment (LF-IQA) methods. In this paper, we propose a No-Reference (NR) LF-IQA method that is based on reference-free distortion maps. With this goal, we first generate a synthetically distorted dataset of 2D images. Then, we compute SSIM distortion maps of these images and use these maps as ground error maps. We train a GAN architecture using these SSIM distortion maps as quality labels. This trained model is used to generate reference-free distortion maps of sub-aperture images of LF contents. Finally, the quality prediction is obtained performing the following steps: 1) perform a non-linear dimensionality reduction with a isometric mapping of the generated distortion maps to obtain the LFI feature vectors and 2) perform a regression using a Random Forest Regressor (RFR) algorithm to obtain the LF quality estimates. Results show that the proposed method is robust and accurate, outperforming several state-of-the-art LF-IQA methods.

Keywords