Efficient Depth Image Reconstruction using Accelerated Proximal Gradient Method

Abstract

This paper presents an accelerated proximal gradient technique for depth image reconstruction from incomplete data samples. One of the most technical challenges in depth imaging is to estimate a depth image from sparse measurements obtained due to missing depth values or fast sensing. Recent advances in signal processing, i.e., compressive sensing (CS) allows the depth image to be reconstructed precisely from far reduced measurements provided that the image has sparse representations in proper bases. Inspired by the CS theory, this paper formulates the task of depth channel reconstruction as a sparsity-regularized least squares optimization problem. To solve this problem efficiently, an iterative algorithm based on the accelerated proximal gradient technique is developed, which not only speeds up the convergence rate, but also enhances the quality of depth image estimation. Several experiments are conducted and the results confirm the efficiency of the proposed approach.