Removing reflection artefacts from a single-image is a problem of both theoretical and practical interest. Removing these artefacts still presents challenges because of the massively ill-posed nature of reflection suppression. In this work, we propose a technique based on a novel optimisation problem. Firstly, we introduce an $H^2$ fidelity term, which preserves fine detail while enforcing global colour similarity. Secondly, we introduce a spatially dependent gradient sparsity prior, which allows user guidance to prevent information loss in reflection-free areas. We show that this combination allows us to mitigate some major drawbacks of the existing methods for reflection removal. We demonstrate, through numerical and visual experiments, that our method is able to outperform the state-of-the-art methods and compete against a recent deep learning approach.