The non-stationary nature of image characteristics calls for adaptiveprocessing, based on the local image content. We propose a simple and flexiblemethod to learn local tuning of parameters in adaptive image processing: weextract simple local features from an image and learn the relation betweenthese features and the optimal filtering parameters. Learning is performed byoptimizing a user defined cost function (any image quality metric) on atraining set. We apply our method to three classical problems (denoising,demosaicing and deblurring) and we show the effectiveness of the learnedparameter modulation strategies. We also show that these strategies areconsistent with theoretical results from the literature.
展开▼
