Epipolar Plane Images from Light Field Images

[Sana Alamgeer]

Hello, I am Sana, a student of PhD and recently started working on Light Field Images. I have explored LightFieldTool to explore different available algorithms. Currently, I am working on a regression problem that has a two-stream CNN. The first stream will process sub-aperture images (SAIs) and the second stream will process Epipolar Plane Images (EPIs) (as shown in the image attached with the email). Now, I am searching for an algorithm to generate EPIs. Unfortunately, the current LightFieldTool has no algorithm to generate EPIs from SAIs of Light Field Images (LFIs). I am writing this email to seek for help to clarify my few confusions. My first question is: Please suggest any algorithm that generates EPIs. (Now that my concepts related to EPI generation is not clear, this has raised the second question below.) My second question is: a - How should I process information of LFIs by using 2-stream CNN? I mean should I process SAI and their corresponding EPIs? or b - Should I take averaged features from all SAIs and EPIs and process the computed averaged feature vector for each LFI. I shall be very grateful for the response. Thank you. -- Best Regards, Sana Alamgeer --------------------------------------------------------------- Universidade de Brasília Departamento de Engenharia Elétrica Campus Universitário Darcy Ribeiro Brasília - DF - Brazil ​