Efficient and Accurate Disparity Estimation from MLA-Based Plenoptic Cameras
This manuscript focuses on the processing images from microlens-array based plenoptic cameras. These cameras enable the capturing of the light field in a single shot, recording a greater amount of information with respect to conventional cameras, allowing to develop a whole new set of applications. However, the enhanced information introduces additional challenges and results in higher computational effort. For one, the image is composed of thousand of micro-lens images, making it an unusual case for standard image processing algorithms. Secondly, the disparity information has to be estimated from those micro-images to create a conventional image and a three-dimensional representation. Therefore, the work in thesis is devoted to analyse and propose methodologies to deal with plenoptic images.
A full framework for plenoptic cameras has been built, including the contributions described in this thesis. A blur-aware calibration method to model a plenoptic camera, an optimization method to accurately select the best microlenses combination, an overview of the different types of plenoptic cameras and their representation. Datasets consisting of both real and synthetic images have been used to create a benchmark for different disparity estimation algorithm and to inspect the behaviour of disparity under different compression rates. A robust depth estimation approach has been developed for light field microscopy and image of biological samples.