This thesis investigates the problem of learning sparse data models and their applications to image processing, regarding both the synthesis and the analysis point of view. Two algorithms called Separable Dictionary Learning (SeDiL) and Geometric Analysis Operator Learning (GOAL) are introduced, which are based on geometric optimization on manifolds. These general models are used to tackle the classical inverse problems of image denoising, inpainting, and superresolution. In addition, an extension of the analysis model is proposed that is able to model dependencies of different signal modalities. Depth map superresolution results based on corresponding depth- and intensity-information manifest the effectiveness of this model.
«
This thesis investigates the problem of learning sparse data models and their applications to image processing, regarding both the synthesis and the analysis point of view. Two algorithms called Separable Dictionary Learning (SeDiL) and Geometric Analysis Operator Learning (GOAL) are introduced, which are based on geometric optimization on manifolds. These general models are used to tackle the classical inverse problems of image denoising, inpainting, and superresolution. In addition, an extensi...
»