Regularizing Image Reconstruction for Gradient-Domain Rendering with Feature Patches

Computer Graphics Forum (Proceedings of Eurographics), 2016

  • Marco Manzi
  • Delio Vicini
  • Matthias Zwicker
teaser figure

Our regularized reconstruction for gradient-domain rendering obtains a high-quality image from a noisy base image, the sampled gradients, and auxiliary features (left). We (right) achieve significantly better images than standard L1 reconstruction (middle). The depicted features are, from left to right, the vertical and horizontal gradients, normals, texture values, positions and ambient occlusion values.


We present a novel algorithm to reconstruct high-quality images from sampled pixels and gradients in gradient-domain rendering. Our approach extends screened Poisson reconstruction by adding additional regularization constraints. Our key idea is to exploit local patches in feature images, which contain per-pixels normals, textures, position, etc., to formulate these constraints. We describe a GPU implementation of our approach that runs on the order of seconds on megapixel images. We demonstrate a significant improvement in image quality over screened Poisson reconstruction under the L1 norm. Because we adapt the regularization constraints to the noise level in the input, our algorithm is consistent and converges to the ground truth.