Gallagher Pryor

3D nonrigid registration via optimal mass transport on the GPU

In this paper, we present a new computationally efficient numerical scheme for the minimizing flow approach for optimal mass transport (OMT) with applications to non-rigid 3D image registration. The approach utilizes all of the gray-scale data in both images, and the optimal mapping from image A to image B is the inverse of the optimal mapping from B to A. Further, no landmarks need to be specified, and the minimizer of the distance functional involved is unique. Our implementation also employs multigrid, and parallel methodologies on a consumer graphics processing unit (GPU) for fast computation. Although computing the optimal map has been shown to be computationally expensive in the past, we show that our approach is orders of magnitude faster then previous work and is capable of finding transport maps with optimality measures (mean curl) previously unattainable by other works (which directly influences the accuracy of registration). We give results where the algorithm was used to compute non-rigid registrations of 3D synthetic data as well as intra-patient pre-operative and post-operative 3D brain MRI datasets.

High-level GPU computing with jacket for MATLAB and C/C++

We describe a software platform for the rapid development of general purpose GPU (GPGPU) computing applications within the MATLAB computing environment, C, and C++: Jacket. Jacket provides thousands of GPU-tuned function syntaxes within MATLAB, C, and C++, including linear algebra, convolutions, reductions, and FFTs as well as signal, image, statistics, and graphics libraries. Additionally, Jacket includes a compiler that translates MATLAB and C++ code to CUDA PTX assembly and OpenGL shaders on demand at runtime. A facility is also included to compile a domain specific version of the MATLAB language to CUDA assembly at build time. Jacket includes the first parallel GPU FOR-loop construction and the first profiler for comparative analysis of CPU and GPU execution times. Jacket provides full GPU compute capability on CUDA hardware and limited, image processing focused compute on OpenGL/ES (2.0 and up) devices for mobile and embedded applications.

Flying in formation using a pursuit guidance algorithm

This paper describes a simple, straightforward algorithm for flying in formation of multiple unmanned air vehicles (UAVs). In particular, we are interested in a formation with no communication between the vehicles. It is assumed that relative kinematics parameters are available to each UAV from an on-board passive sensor followed by estimation processes and a controller which may use visual information. The paper introduces a guidance algorithm, which is based on the theory of pursuit curves in conjunction with a velocity controller.

Knowledge-Based Segmentation for Tracking Through Deep Turbulence

A combined knowledge-based segmentation/active contour algorithm is used for target tracking through turbulence. The algorithm utilizes Bayesian modeling for segmentation of noisy imagery obtained through longrange, laser imaging of a distance target, and active contours for tip tracking. The algorithm demonstrates improved target tracking performance when compared to weighted centroiding. Open-loop and closed-loop comparisons of the algorithms using simulated imagery validate the hypothesis.

Fast Optimal Mass Transport for Dynamic Active Contour Tracking on the GPU

In computational vision, visual tracking remains one of the most challenging problems due to noise, clutter, occlusion, and dynamic scenes. No one technique has yet managed to solve this problem completely, but those that employ control- theoretic filtering techniques have proven to be quite successful. In this work, we extend one such technique by Niethammer et al. in which implicitly represented dynamically evolving contours are filtered using a geometric observer framework. The effectiveness of the observer hangs upon the solution of two major problems: (1) the calculation of accurate curve velocities and (2) the determination of diffeomorphic correspondence maps between curves for geometric interpolation. We propose the use of novel image registration techniques such as image warping and optimal mass transport for the solution of these problems which increase the performance of the framework and reduce algorithmic complexity. One major drawback to the original scheme, as it relies on PDE solutions, is its computational burden restricting it from real time use. We show that the framework can, in fact, run in near real time by implementing our additions to the framework on the graphics processing unit (GPU) and show better execution times for these algorithms than reported in recent literature.

Layered Active Contours for Tracking

Presented at British Machine Vision Conference 2007, University of Warwick, UK, September 10-13, 2007.

Jacket: GPU Powered MATLAB Acceleration

Publisher Summary Jacket is a software platform developed at AccelerEyes, which allows users to execute MATrix LABoratory (MATLAB) M-code on CUDA-capable graphics processing units (GPUs). MATLAB by the MathWorks is a standard platform for technical computing and graphics in science, engineering, and finance. The combination of a simple matrix language, interactive prompt, automatic memory management, and on-the-fly compilation make MATLAB well suited to rapid prototyping of algorithms and exploring data. MATLABs one drawback is performance, and Jacket alleviates this by seamless offloading of computations to the GPU. Jacket provides users access to a set of libraries, functions, and tools that facilitate numerical computation on the GPU including multi-GPU support built on MATLABs Parallel Computing Toolbox and Distributed Computing Server. Jacket has been designed for programmers who have large data-parallel tasks but who are not low level programmers accustomed to dealing with GPU-specific constructs. Once data is marked as “GPU” data using these functions, Jacket provides native GPU implementations of a large set of the standard MATLAB functions to operate on that data. Jacket achieves transparency by defining a new set of classes dubbed “g” objects, where each element of this set corresponds to a base class of the MATLAB standard interface—single, uint16, ones, etc. map to gsingle, guint16, gones, etc. The Jacket architecture uses object-oriented programming to handle references to data. Jacket includes a graphics toolbox that provides a simple method of displaying computational results on the GPU without bringing those results back to the host.

Fast Multigrid Optimal Mass Transport for Image Registration and Morphing.

In this paper we present a novel, computationally efficient algorithm for nonrigid 2D image registration based on the work of Haker et al.[1, 2]. We formulate the registration task as an Optimal Mass Transport (OMT) problem based on the Monge-Kantorovich theory. This approach gives a number of advantages over other conventional registration methods: (1) It is parameter free and no landmarks need to be specified, (2) it is symmetrical and the energy functional has a unique minimiser, and (3) it can register images where brightness constancy is an invalid assumption. Our algorithm solves the Optimal Mass Transport program via multi-resolution, multi-grid, and parallel methodologies on a consumer graphics processing unit (GPU). Although solving the OMT problem has been shown to be computationally expensive in the past, we show that our approach is almost two orders magnitude faster than previous work and is capable of finding transport maps with optimality measures (mean curl) previously unattainable by other works (which directly influences the quality of registration). We give results where the algorithm was used to register 2D short axis cardiac MRI images and to morph two image sets from a SOHO solar flare image sequence.