Computationally Efficient Formulation of Sparse Color Image Recovery in the JPEG Compressed Domain

摘要: Sparse representations provide a powerful framework for various image processing tasks, among which recovery seems to be an already classical application. While most developments of applications are focused on finding the best dictionary, possibility using existing sparse tends ignored. This is case JPEG compressed representation, representation in terms discrete cosine transform (DCT) dictionary. The development frameworks directly encoded can lead computationally efficient approaches. Here we introduce DCT-based domain formulation color process within and prove mathematically experimentally not only its numerical efficiency as compared pixel level (the time reduced up 40 %), but also good quality restoration results.