Abstract: With known image upscaling methods of noisy images, important detail information is lost during denoising. A method for upscaling noisy input images comprises upscaling a noisy input image to obtain a noisy High-Resolution (HR) image, denoising the noisy input image to obtained a denoised Low-Resolution (LR) image, upscaling the denoised LR image to obtain an upscaled denoised LR image, and combining the noisy HR image and the upscaled denoised LR image to generate a denoised HR image.

Abstract: A noise-aware single-image super-resolution (SI-SR) method and apparatus automatically cancels additive noise while adding detail learned from lower scale of an input image. A recent and efficient in-place cross-scale self-similarity prior is exploited for both learning fine detail examples to complement the interpolation-based upscaled image patches and reducing image noise.

Abstract: Image super-resolution (SR) generally enhance the resolution of images. One of SR's main challenge is discovering mappings between low-resolution (LR) and high-resolution (HR) image patches. The invention learns patch upscaling projection matrices from a training set of images. Input images are divided into overlapping patches, which are normalized and transformed to a defined orientation. Different transformations can be recognized and dealt with by using a simple 2D-projection. The transformed patches are clustered, and cluster specific upscaling projection matrices and corresponding cluster centroids determined during training are applied to obtain upscaled patches. The upscaled patches are assembled to an upscaled image.

Abstract: Known super-resolution algorithms are inefficient due to a high degree of redundant calculations, require search operations such as block matching for finding nearest neighbors, and achieve only small magnification factors. An improved method for upscaling an image comprises steps of upscaling the image by pixel interpolation to obtain a coarse upscaled image, and, for each pixel of the coarse upscaled image, determining a nonlinear regression function for a patch of pixels around a current pixel of the coarse upscaled image and enhancing the value of the current pixel by adding the result of the nonlinear regression function, wherein a pixel of an upscaled image is obtained. The nonlinear regression function is obtained from a trained regression tree, based on geometric features of the patch.

Abstract: An algorithm for performing super-resolution splits an input image or video into patches and relies on image self-similarity, wherein similar patches are searched in different downscaled versions of an image, using Approximate Nearest-Neighbor Fields (ANNF). The goal of ANNF is to locate with a minimal number of search iterations for each patch of a source image the k most similar patches in a downscaled version of the source image or video. A method for generating an ANNF for images of an input video comprises generating a plurality of downscaled versions of the images of the input video at different scales, generating an Inverse ANNF (IANNF) for the input video by finding for each patch of the downscaled images similar patches in the input video, generating an ANNF for the input video by reversing the IANNF, and filling gaps in the ANNF by random search.

Abstract: Known methods for generating super-resolved images from single input images have various disadvantages. An improved method for generating a super-resolved image from a single low-resolution input image comprises up-scaling the input image to generate an initial version of the super-resolved image, searching, for each patch of the low-resolution input image, similar low-resolution patches in first search windows within down-sampled versions of the input image, and determining, in less down-sampled versions of the input image, high-resolution patches that correspond to the similar low-resolution patches. The determined high-resolution patches are cropped, a second search window is determined in the initial version of the super-resolved image, and a best-matching position for each cropped high-resolution patch is searched within the second search window. Finally, each cropped high-resolution patch is added to the super-resolved image at its respective best-matching position.

Abstract: A method and an apparatus (20) for up-scaling an input image (12) are described, wherein a cross-scale self-similarity matching using superpixels is employed to obtain substitutes for missing details in an up-scaled image. The apparatus (20) comprises a superpixel vector generator (7) configured to generate (10) consistent superpixels for the input image (12) and one or more auxiliary input images (I1, I3) and to generate (11) superpixel test vectors based on the consistent superpixels. A matching block (5) performs a cross-scale self-similarity matching (12) across the input image (12) and the one or more auxiliary input images (I1, I3) using the superpixel test vectors. Finally, an output image generator (22) generates (13) an up-scaled output image (O2) using results of the cross-scale self-similarity matching (12).

Abstract: A noise-aware single-image super-resolution (SI-SR) method and apparatus automatically cancels additive noise while adding detail learnt from lower scale of an input image. In contrast to common SI-SR techniques, the input image is not necessarily assumed to be a clean source of examples. Instead, a recent and efficient in-place cross-scale self-similarity prior is exploited for both learning fine detail examples to complement the interpolation-based upscaled image patches and reducing image noise. Experimental results show a promising performance, despite of the relatively simple algorithm. Both objective and subjective evaluations show large quality improvements when upscaling images immersed in noise.

Abstract: Image super-resolution (SR) generally enhance the resolution of images. One of SR's main challenge is discovering mappings between low-resolution (LR) and high-resolution (HR) image patches. The invention learns patch upscaling projection matrices from a training set of images. Input images are divided into overlapping patches, which are normalized and transformed to a defined orientation. Different transformations can be recognized and dealt with by using a simple 2D-projection. The transformed patches are clustered, and cluster specific upscaling projection matrices and corresponding cluster centroids determined during training are applied to obtain upscaled patches. The upscaled patches are assembled to an upscaled image.

Abstract: Traditional image denoising requires image analysis or noise level analysis. Differently, the invention provides denoising by dividing an input image into small square overlapping patches, computing and storing the mean value of each patch, and subtracting it from the patch. Each zero-mean patch is then automatically aligned to a reference orientation by computing a few relevant DCT coefficients of the patch, analyzing the patch orientation in terms of transposition, inversion and horizontal and vertical flipping, and applying a re-orientation transform to automatically pose the patch in a standard orientation, regardless of its contents. The reoriented patches are clustered, and each of the resulting clusters is either shuffled or averaged. Then, all patches are re-transformed back to their original orientations by reversing the previous transforms, their respective mean is added and the denoised image is reconstructed by overlapping the re-transformed and mean added patches.

Abstract: A method for performing super-resolution on an input image having low resolution, comprises generating a generic training data set of descriptors extracted from regions of training images, and for each patch of the input image, determining a defined number of nearest neighbor regions, extracting example patches from the nearest neighbor regions and collecting the example patches in an example patch data base, determining a combination of low-resolution example patches that, according to their descriptors, optimally approximate the current patch, and constructing a high-resolution patch, wherein a super-resolved image is obtained.

Abstract: A method and an apparatus for improving a main image by fusing the richer information contained in a secondary image are described. A 3D structure of objects contained in the secondary image is retrieved and a parallax-corrected version of the secondary image is generated using the 3D structure. For this purpose a camera pose for which a projection of the 3D structure of the objects contained in the secondary image best resembles the perspective in the main image is determined and the parallax-corrected version of the secondary image is synthesized based on the determined camera pose. The parallax-corrected version of the secondary image is then fused with the main image.

Abstract: Known super-resolution algorithms are inefficient due to a high degree of redundant calculations, require search operations such as block matching for finding nearest neighbors, and achieve only small magnification factors. An improved method for upscaling an image comprises steps of upscaling the image by pixel interpolation to obtain a coarse upscaled image, and, for each pixel of the coarse upscaled image, determining a nonlinear regression function for a patch of pixels around a current pixel of the coarse upscaled image and enhancing the value of the current pixel by adding the result of the nonlinear regression function, wherein a pixel of an upscaled image is obtained. The nonlinear regression function is obtained from a trained regression tree, based on geometric features of the patch.

Abstract: A method for performing super-resolution on an input image having low resolution, comprises generating a generic training data set of descriptors extracted from regions of training images, and for each patch of the input image, determining a defined number of nearest neighbor regions, extracting example patches from the nearest neighbor regions and collecting the example patches in an example patch data base, determining a combination of low-resolution example patches that, according to their descriptors, optimally approximate the current patch, and constructing a high-resolution patch, wherein a super-resolved image is obtained.

Abstract: Known methods for generating super-resolved images from single input images have various disadvantages. An improved method for generating a super-resolved image from a single low-resolution input image comprises up-scaling the input image to generate an initial version of the super-resolved image, searching, for each patch of the low-resolution input image, similar low-resolution patches in first search windows within down-sampled versions of the input image, and determining, in less down-sampled versions of the input image, high-resolution patches that correspond to the similar low-resolution patches. The determined high-resolution patches are cropped, a second search window is determined in the initial version of the super-resolved image, and a best-matching position for each cropped high-resolution patch is searched within the second search window. Finally, each cropped high-resolution patch is added to the super-resolved image at its respective best-matching position.

Abstract: In a method for performing super-resolution of a single image, a high-resolution version of an observed image is generated by exploiting cross-scale self-similarity, wherein up-scaling and analysis filters are used. The up-scaling and analysis filters are adaptively selected according to local kernel cost.

Abstract: An algorithm for performing super-resolution splits an input image or video into patches and relies on image self-similarity, wherein similar patches are searched in different downscaled versions of an image, using Approximate Nearest-Neighbor Fields (ANNF). The goal of ANNF is to locate with a minimal number of search iterations for each patch of a source image the k most similar patches in a downscaled version of the source image or video. A method for generating an ANNF for images of an input video comprises generating a plurality of downscaled versions of the images of the input video at different scales, generating an Inverse ANNF (IANNF) for the input video by finding for each patch of the downscaled images similar patches in the input video, generating an ANNF for the input video by reversing the IANNF, and filling gaps in the ANNF by random search.

Abstract: A method and an apparatus for improving a main image by fusing the richer information contained in a secondary image are described. A 3D structure of objects contained in the secondary image is retrieved and a parallax-corrected version of the secondary image is generated using the 3D structure. For this purpose a camera pose for which a projection of the 3D structure of the objects contained in the secondary image best resembles the perspective in the main image is determined and the parallax-corrected version of the secondary image is synthesized based on the determined camera pose. The parallax-corrected version of the secondary image is then fused with the main image.