Integrating wavelet transformation with Markov random field analysis for the depth estimation of light-field images

This study addresses the problem of recovering the three-dimensional depth data from the images taken by a light-field camera. Unlike the conventional approach to extract the depth information from the spatial and the angular gradients in the epipolar plane images (EPIs), this study proposes to check the similarity between the pixels for the estimation of EPI slopes and use a wavelet transformation augmented multi-scale analysis to perform smart segmentations. The Markov random field is then applied for surface reconstruction. The proposed algorithm offers significant improvement to the depth estimation, especially for noise contaminated images. Application of the method on the light-field images and on synthesised data show that the proposed method is robust against the noise and achieves better estimation results compared with the available literature.