An image super-resolution method from multiple observation of low-resolutionimages is proposed. The method is based on sub-pixel accuracy block matchingfor estimating relative displacements of observed images, and sparse signalrepresentation for estimating the corresponding high-resolution image. Relativedisplacements of small patches of observed low-resolution images are accuratelyestimated by a computationally efficient block matching method. Since theestimated displacements are also regarded as a warping component of imagedegradation process, the matching results are directly utilized to generatelow-resolution dictionary for sparse image representation. The matching scoresof the block matching are used to select a subset of low-resolution patches forreconstructing a high-resolution patch, that is, an adaptive selection ofinformative low-resolution images is realized. When there is only onelow-resolution image, the proposed method works as a single-framesuper-resolution method. The proposed method is shown to perform comparable orsuperior to conventional single- and multi-frame super-resolution methodsthrough experiments using various real-world datasets.
展开▼
