Abstract: A method generates a high-resolution (HR) image from a low-resolution (LR) image using regression functions. During a training stage, training HR images are downsampled to LR images. A signature is determined for each LR-HR patch pair based on a local ternary pattern (LTP). The signature is a low dimensional descriptor used as an abstraction of the patch pair features. Then, patch pairs with the same signature are clustered, and a regression function which maps the LR patches to the HR patches is determined. In some cases patch pairs of similar signatures can be combined for learning and a single regression function determined, thus decreasing the number of required regression functions. During actual upscaling, LR patches of an input image are similarly processed to obtain the signatures and from the regression functions. The LR patches can then be upscaled using the training regression functions.

Abstract: A method detects an object in a scene by first determining an active set of window positions from depth data. Specifically, the object can be a person. The depth data are acquired by a depth sensor. For each, window position perform the following steps. Assign a window size based on the depth data. Select a current window from the active set of window positions. Extract a joint feature from the depth data and texture data for the current window, wherein the texture data are acquired by a camera. Classify the joint feature to detect the object. The classifier is trained with joint training features extracted from training data including training depth data and training texture data acquired by the sensor and camera respectively. Finally, the active set of window positions is updated before processing the next current window.

Abstract: Three-dimensional (3D) geometric boundaries are detected in images of a scene that undergoes varying lighting conditions caused by light sources in different positions, from a set of input images of the scene illuminated by at least two different lighting conditions. The images are aligned, e.g., acquired by a stationary camera, so that pixels at the same location in all of the input images correspond to the same point in the scene. For each location, a patch of corresponding pixels centered at the location is extracted from each input image. For each location, a confidence value that there is a 3D geometric boundary at the location is determined.

Abstract: A set of nonnegative lighting basis images representing a scene illuminated by a set of stationary light sources is recovered from a set of input images of the scene that were acquired by a stationary camera. Each image is illuminated by a combination of the light sources, and at least two images in the set are illuminated by different combinations. The set of input images is factorized into the nonnegative lighting basis images and a set of indicator coefficients, wherein each lighting basis image corresponds to an appearance of the scene illuminated by one of the light sources, and wherein each indicator coefficient indicates a contribution of one of the light sources to one of the input images.

Abstract: An input image is denoised by first constructing a pixel-wise noise variance map from the input image. The noise has spatially varying variances. The input image is partitioned into patches using the noise variance map. An intermediate image is determined from the patches. Collaborative filtering is applied to each patch in the intermediate image using the noise variance map to produce filtered patches. Then, the filtered patches are projected to an output image.

Abstract: A method generates a high-resolution (HR) image from a low-resolution (LR) image using regression functions. During a training stage, training HR images are downsampled to LR images. A signature is determined for each LR-HR patch pair based on a local ternary pattern (LTP). The signature is a low dimensional descriptor used as an abstraction of the patch pair features. Then, patch pairs with the same signature are clustered, and a regression function which maps the LR patches to the HR patches is determined. In some cases patch pairs of similar signatures can be combined for learning and a single regression function determined, thus decreasing the number of required regression functions. During actual upscaling, LR patches of an input image are similarly processed to obtain the signatures and from the regression functions. The LR patches can then be upscaled using the training regression functions.

Abstract: A tumor is tracked in multiple sequences of images acquired concurrently from different viewpoints. Features are extracted in each set of current images using a window. A regression function, subject to motion constraints, is applied to the features to obtain 3D motion parameters, which are applied to the tumor as observed in the images to obtain a 3D location of the object. Then, the shape of the 3D object at the 3D location is projected onto each image to update the location of the window for the next set of images to be processed.

Abstract: Four-dimensional (4D) computed tomography (CT) is simulated by first generating a surface mesh from a single thoracic CT scan. Tetrahedralization is applied to the surface mesh to obtain a first volume mesh. Finite element analysis, using boundary constraints and load definitions, is applied to the first volume mesh to obtain a lung deformation according to an Ogden model. Constrained tetrahedralization, using control points, is applied to the lung deformation to obtain a second volume mesh, which is then deformed using mass-spring-damper simulation to produces the 4DCT.

Abstract: A dictionary of atoms for coding data is learned by first selecting samples from a set of samples. Similar atoms in the dictionary are clustered, and if a cluster has multiple atoms, the atoms in that cluster are merged into a single atom. The samples can be acquired online.

Abstract: A method segments n-dimensional by first determining prior information from the data. A fidelity term is determined from the prior information, and the data are represented as a graph. A graph Laplacian is determined from the graph from the graph, and a Laplacian spectrum constraint is determined from the graph Laplacian. Then, an objective function is minimized according to the fidelity term and the Laplacian spectrum constraint to identify a segment of target points in the data.

Abstract: Blocking artifacts are reduced by projecting each patch obtained from an input image onto a set of bases vectors to determine multiple representations for each patch. The set of bases vectors are learned from a training image, and the bases vectors include a full basis vector, and one or two subspace bases vectors. An optimal basis vector is determined in the set of bases vectors for each patch according to the projection. A threshold is applied to coefficients of the optimal basis vector to determine a filtered representation for each patch, and a reconstructed patch is generated using the filtered representation. Then, the aggregating the reconstructed patches are aggregated to produce an output image.

Abstract: A method recovers an uncorrupted low-rank matrix, noise in corrupted data and a subspace from the data in a form of a high-dimensional matrix. An objective function minimizes the noise to solve for the low-rank matrix and the subspace without estimating the rank of the low-rank matrix. The method uses group sparsity and the subspace is orthogonal. Random subsampling of the data can recover subspace bases and their coefficients from a much smaller matrix to improve performance. Convergence efficiency can also be improved by applying an augmented Lagrange multiplier, and an alternating stepwise coordinate descent. The Lagrange function is solved by an alternating direction method.

Abstract: A method detects an object in a scene by first determining an active set of window positions from depth data. Specifically, the object can be a person. The depth data are acquired by a depth sensor. For each window position perform the following steps. Assign a window size based on the depth data. Select, a current window from the active set of window positions. Extract a joint feature from the depth data and texture data for the current window, wherein the texture data are acquired by a camera. Classify the joint feature to detect the object. The classifier is trained with joint training features extracted from training data including training depth data and training texture data acquired by the sensor and camera respectively. Finally, the active set of windows position is updated before processing the next current window.

Abstract: A set of nonnegative lighting basis images representing a scene illuminated by a set of stationary light sources is recovered from a set of input images of the scene that were acquired by a stationary camera. Each image is illuminated by a combination of the light sources, and at least two images in the set are illuminated by different combinations. The set of input images is decomposed into the nonnegative lighting basis images and a set of indicator coefficients, wherein each lighting basis image corresponds to an appearance of the scene illuminated by one of the light sources, and wherein each indicator coefficient indicates a contribution of one of the light sources to one of the input images.

Abstract: Three-dimensional (3D) geometric boundaries are detected in images of a scene that undergoes varying lighting conditions caused by light sources in different positions, from a set of input images of the scene illuminated by at least two different lighting conditions. The images are aligned, e.g., acquired by a stationary camera, so that pixels at the same location in all of the input images correspond to the same point in the scene. For each location, a patch of corresponding pixels centered at the location is extracted from each input image. For each location, a confidence value that there is a 3D geometric boundary at the location is determined.

Abstract: An input image is denoised by first constructing a pixel-wise noise variance map from the input image. The noise has spatially varying variances. The input image is partitioned into patches using the noise variance map. An intermediate image is determined from the patches. Collaborative filtering is applied to each patch in the intermediate image using the noise variance map to produce filtered patches. Then, the filtered patches are projected to an output image.

Abstract: A shape of an object is represented by a set of points inside and outside the shape. A decision function is learned from the set of points an object. Feature points in the set of points are selected using the decision function, or a gradient of the decision function, and then a local descriptor is determined for each feature point.

Abstract: A tumor is tracked in multiple sequences of images acquired concurrently from different viewpoints. Features are extracted in each set of current images using a window. A regression function, subject to motion constraints, is applied to the features to obtain 3D motion parameters, which are applied to the tumor as observed in the images to obtain a 3D location of the object. Then, the shape of the 3D object at the 3D location is projected onto each image to update the location of the window for the next set of images to be processed.

Abstract: A natural input image is upscaled, first by interpolation. Second, edges in the interpolated image are sharpened by a lion-parametric patch transform. The result is decomposed into an edge layer and a detail layer. Only pixels in the detail layer enhanced, and the enhanced detail layer is merged with the edge layer to produce a high resolution version of the input image.

Abstract: Blocking artifacts are reduced by projecting each patch obtained from an input image onto a set of bases vectors to determine multiple representations for each patch. The set of bases vectors are learned from a training image, and the bases vectors include a full basis vector, and one or two subspace bases vectors. An optimal basis vector is determined in the set of bases vectors for each patch according to the projection. A threshold is applied to coefficients of the optimal basis vector to determine a filtered representation for each patch, and a reconstructed patch is generated using the filtered representation. Then, the aggregating the reconstructed patches are aggregated to produce an output image.