No abstract

No abstract

Abstract-Local learning of sparse image models has proven to be very effective to solve inverse problems in many computer vision applications. To learn such models, the data samples are often clustered using the K-means algorithm with the Euclidean distance as a dissimilarity metric. However, the Euclidean distance may not always be a good dissimilarity measure for comparing data samples lying on a manifold. In this paper, we propose two algorithms for determining a local subset of training samples from which a good local model can be computed for reconstructing a given input test sample, where we take into account the underlying geometry of the data. The first algorithm, called Adaptive Geometry-driven Nearest Neighbor search (AGNN), is an adaptive scheme which can be seen as an out-of-sample extension of the replicator graph clustering method for local model learning. The second method, called Geometrydriven Overlapping Clusters (GOC), is a less complex nonadaptive alternative for training subset selection. The proposed AGNN and GOC methods are evaluated in image super-resolution, deblurring and denoising applications and shown to outperform spectral clustering, soft clustering, and geodesic distance based subset selection in most settings.

A: Convolutional neural networks (CNNs) have found applications in many image processing tasks, such as feature extraction, image classification, and object recognition. It has also been shown that the inverse of CNNs, so-called deconvolutional neural networks, can be used for inverse problems such as plasma tomography. In essence, plasma tomography consists in reconstructing the 2D plasma profile on a poloidal cross-section of a fusion device, based on line-integrated measurements from multiple radiation detectors. Since the reconstruction process is computationally intensive, a deconvolutional neural network trained to produce the same results will yield a significant computational speedup, at the expense of a small error which can be assessed using different metrics. In this work, we discuss the design principles behind such networks, including the use of multiple layers, how they can be stacked, and how their dimensions can be tuned according to the number of detectors and the desired tomographic resolution for a given fusion device. We describe the application of such networks at JET and COMPASS, where at JET we use the bolometer system, and at COMPASS we use the soft X-ray diagnostic based on photodiode arrays. K : Computerized Tomography (CT) and Computed Radiography (CR); Plasma diagnostics -interferometry, spectroscopy and imaging 1Corresponding author. 2See the author list of Overview of the JET preparation for Deuterium-Tritium Operation by E. Joffrin et al. in Nucl.

This paper describes a new single-image super-resolution algorithm based on sparse representations with image structure constraints. A structure tensor based regularization is introduced in the sparse approximation in order to improve the sharpness of edges. The new formulation allows reducing the ringing artefacts which can be observed around edges reconstructed by existing methods. The proposed method, named Sharper Edges based Adaptive Sparse Domain Selection (SE-ASDS), achieves much better results than many stateof-the-art algorithms, showing significant improvements in terms of PSNR (average of 29.63, previously 29.19), SSIM (average of 0.8559, previously 0.8471) and visual quality perception.