Projecte final de carrera fet en col.laboració amb Illinois Institute of Technology

English: Digital image restoration plays a very important role in many fields such as surveillance or medical imaging, where it can be used to obtain high-resolution images so that a more accurate analysis can be performed. In this study we firstly introduce the reader to classical image restoration techniques, such as the Inverse filter or the Wiener filter. However, the main objective of this study is to evaluate a new approach to digital image restoration, which is based on a mesh model of the image. In order to create the mesh model, the digital image is non-uniformly sampled with the use of an algorithm based on a feature map of the image and the classical Floyd-Steinberg error-diffusion. As a result, the sampling is adapted to the content of the image, so more samples are placed in areas with more image detail (highfrequency areas) and less samples are placed in smooth regions (low-frequency areas). The samples (also called mesh nodes) are then connected using the Delaunay triangulation algorithm in order to form the mesh structure. An iterative least-squares fitting algorithm is then used to calculate the intensity of the mesh elements in order to obtain an accurate approximation of the image. The proposed method is an effective image restoration technique for digital images degraded by blur and noise. Moreover, the use of a content-adaptive mesh model (CAMM) means a compression of the image, because less samples are needed in order to represent the image since they are adapted to the image features. The results obtained demonstrate that the adaptive mesh model method can outperform the classical image restoration techniques presented in this study