UNC Image-Based Rendering Research

Inverse Warper

In forward warping the pixels in the reference image are transfered in the desired image by computing their new coordinates; this often leads to a non uniform distribution of "known pixels" in the desired image, making it hard to reconstruct.

The inverse warping aproach is based on finding for every pixel in the desired image the corresponding pixel from the reference image. This will lead to a better desired image at least by exposing the reconstruction problems. This is achieved with the price of a slower algorithm since the coordinates of the corresponding pixel in the reference image are determined at the end of a search of running time proportional to the bigger dimension of the rectangular reference image.

Here is my C++ implementation of an inverse warper based on section 7.2 of Leonard McMillan's doctoral dissertation, CS Dept, UNC at CH, 1997: to code directory