Hanjie Pan
Audiovisual Communications Laboratory (LCAV) at EPFL.
Hanjie Pan
EPFL-IC-LCAV
BC 322 (Bâtiment BC)
Station 14
1015 Lausanne
Image up-sampling aims at reconstructing a high resolution image from a single low resolution one. It is essential to exploit prior knowledge on the reconstructed image in order to better condition this severely ill-conditioned inverse problem. We present a novel edge modelling framework based on a spatial domain interpretation of annihilation of curves with finite rate of innovation. More specifically, we define a continuous domain mask function that vanishes around large gradients, i.e., around edges. The mask function is reconstructed by minimising its product with the gradient image. We show that accurate edge models are reconstructed by assuming a simple local linear edge model. Based on the same idea, we further combine these local edge models and build a global one, which serves as an edge-preserving constraint in image up-samplings. Moreover, we propose an efficient alternating direction method of multipliers (ADMM) to solve the up-sampling problem numerically. Experiments with natural images demonstrate the effectiveness of the global edge model in improving the quality of the up-sampled images thus achieving state of the art performance.
This repository contains the code to reproduce the results of Chapter 6 of Looking beyond Pixels Theory, Algorithms and Applications of Continuous Sparse Recovery. It contains a Matlab implementation of the proposed algorithm.
% image up-sampling with known ground-truth
main_upSamp.m
% blind image up-sampling
main_blindUpSamp.m
Copyright (c) 2018, Hanjie Pan
The source code is released under the MIT license.