<p>Image restoration in the presence of noise and missing data (inpainting) remains a fundamental challenge in image processing. Traditional first-order variational models, while effective, often suffer from undesirable effects such as block artifacts and directional bias. In this paper, we propose a novel first-order total variation model based on a two-stage framework. First, the image is upsampled using bilinear interpolation. Then, two types of triangular meshes are constructed over the expanded domain, and a discrete total variation is defined by averaging contributions from each triangulation. This leads to a geometry-aware variational model that is applied to both denoising and inpainting tasks. Numerical results demonstrate that our method effectively reduces block artifacts and outperforms classical TV, second-order TGV, second-order Shannon TGV, and learning-based methods in terms of visual quality and restoration accuracy.</p>