Optimization-based approaches for image deblurring and denoising on Graphics Processing Units (GPU) are considered. In particular, a new GPU implementation of a recent gradient projection method for edge-preserving removal of Poisson noise is presented. The speedups over standard CPU implementations are evaluated on both synthetic data and astronomical and medical imaging problems.

This research examines methods of implying deconvolution to blurry barcode signals with noise. Our goal is to to take these signals and recontrust them, using Yu Mao’s method of Gradient Projection, to be as clear as possible. This research examines the work of Yu Mao [5], alumni from the University of Minnesota. Our research is motivated by Yu Mao’s findings for how to reconstruct binary funct...

This paper presents a new dual formulation for quadratically constrained convex programs (QCCP). The special structure of the derived dual problem allows to apply the gradient projection algorithm to produce a simple explicit method involving only elementary vector-matrix operations, that is proven to converge at a linear rate.

Compressed sensing is a novel signal sampling theory under the condition that the signal is sparse or compressible. The existing recovery algorithms based on the gradient projection can either need prior knowledge or recovery the signal poorly. In this paper, a new algorithm based on gradient projection is proposed, which is referred as Quasi Gradient Projection. The algorithm presented quasi g...

Although many variants of stochastic gradient descent have been proposed for large-scale convex optimization, most of them require projecting the solution at each iteration to ensure that the obtained solution stays within the feasible domain. For complex domains (e.g., positive semidefinite cone), the projection step can be computationally expensive, making stochastic gradient descent unattrac...

A potentially large anatomical variability among subjects in a population makes nonrigid image registration techniques prone to inaccuracies and to high computational costs in their optimisation. In this paper, we propose a new learning-based approach to accelerate the convergence rate of any chosen parametric energy-based image registration method. From a set of training images and their corre...

Nuisance attribute projection (NAP) was an effective method to reduce session variability in SVM-based speaker verification systems. As the expanded feature space of nonlinear kernels is usually high or infinite dimensional, it is difficult to find nuisance directions via conventional eigenvalue analysis and to do projection directly in the feature space. In this paper, two different approaches...

This paper presents an overview of the issues associated with applying a domain-decomposition message-passing paradigm to the parallel implementation of both explicit and semi-implicit projection algorithms. The use of an element-based domain decomposition with an e cient solution strategy for the pressure eld is shown to yield a scalable, parallel solution method capable of treating complex ow...

This paper presents an overview of the issues associated with applying a domain-decomposition message-passing paradigm to the parallel implementation of both explicit and semi-implicit projection algorithms. The use of an element-based domain decomposition with an eecient solution strategy for the pressure eld is shown to yield a scalable, parallel solution method capable of treating complex ow...