Image Transformation Based on Learning Dictionaries across Image Spaces

摘要: In this paper, we propose a framework of transforming images from source image space to target space, based on learning coupled dictionaries training set paired images. The can be used for applications such as super-resolution and estimation intrinsic components (shading albedo). It is local parametric regression approach, using sparse feature representations over learned across the spaces. After dictionary learning, coefficient vectors patch pairs are partitioned into easily retrievable clusters. For any test patch, fast index its closest cluster perform between obtained representation (together with dictionary) provides multiple constraints each pixel estimated. final reconstructed these constraints. contributions our proposed three-fold. 1) We concept coding which requires pair corresponding patches have same support, i.e., indices nonzero elements. 2) devise partitioning scheme divide high-dimensional but facilitates extremely retrieval clusters query patches. 3) Benefiting feature-based transformation, method more robust corrupted input data, considered simultaneous restoration transformation process. Experiments demonstrate effectiveness efficiency method.