Video editing in the gradient domain using a wavelet based 3-D reconstruction algorithm and an iterative Poisson solver

Abstract

A new wavelet based algorithm for reconstructing three dimensional (3-D) signals from gradients is proposed. The algorithm is based on obtaining directly from the gradients the Haar wavelet decomposition and from it the 3-D signal using a wavelet synthesis that includes an iterative Poisson solver at each resolution. The approach is an extension of a similar approach for wave-front reconstruction in [1]. Experiments with video sequences demonstrate that the proposed algorithm leads to good quality reconstructions in the presence of noise in the gradients. This makes the proposed algorithm valuable for gradient based video processing applications and a video editing example is included to illustrate this.