Jamshed N. Patel

Scalability of 2-D wavelet transform algorithms: analytical and experimental results on coarse-grained parallel computers

We present analytical and experimental results for the scalability of 2-D discrete wavelet transform algorithms on coarse-grained parallel architectures. The principal operation in the 2-D DWT is the filtering operation used to implement the filter banks of the 2-D subband decomposition. We derive analytical results comparing time domain and frequency domain parallel algorithms for realizing the filter banks. Experiments on the Intel Paragon validate the analytical results. We demonstrate that there exist combinations of the machine size, image size, and wavelet size for which the time-domain algorithms outperform the frequency domain algorithms, and vice-versa.

A library-based program development environment for parallel image processing

Cloner is an image processing prototyping environment that helps users design new parallel image processing algorithms for a target machine by building on and modifying existing library algorithms. In this paper we show the Cloner user interface, discuss how guided access is accomplished, and provide an example of how Cloner supports the rapid development of high performance codes. The example demonstrates how menu options and queries are used to guide a user to select an appropriate 2-dimensional FFT algorithm based on image site and available machine resources.<<ETX>>

Scalable parallel implementations of list ranking on fine-grained machines

We present analytical and experimental results for fine-grained list ranking algorithms. We compare the scalability of two representative algorithms on random lists, then address the question of how the locality properties of image edge lists can be used to improve the performance of this highly data-dependent operation. Starting with Wyllies algorithm and Anderson and Millers randomized algorithm as bases, we use the spatial locality of edge links to derive scalable algorithms designed to exploit the characteristics of image edges. Tested on actual and synthetic edge data, this approach achieves significant speedup on the MasPar MP-1 and MP-2, compared to the standard list ranking algorithms. The modified algorithms exhibit good scalability and are robust across a wide variety of image types. We also show that load balancing on fine grained machines performs well only for large problem to machine size ratios.

On the scalability of 2-D wavelet transform algorithms on fine-grained parallel machines

We study the scalability of 2-D discrete wavelet transform algorithms on fine-grained parallel architectures. The principal operation in the 2-D DWT is the filtering operation used to implement the filter banks of the 2-D subband decomposition. We demonstrate that there exist combinations of the machine size, image size, and wavelet size for which the time-domain algorithms outperform the frequency domain algorithms, and vice-versa. We, therefore, demonstrate that a hybrid approach which combines time- and frequency-domain approaches can yield optimal performance for a broad range of problem and machine sizes. Furthermore, we show the effect of processor speed and the use of separable versus nonseparable wavelets on the crossover points between the algorithm approaches.

Contour ranking on coarse grained machines: A case study for low-level vision computations

In this paper we present parallel solutions for performing image contour ranking on coarse-grained machines. In contour ranking, a linear representations of the edge contours is generated from the raw image. We describe solutions that employ different divide-andconquer approaches and that use different communication patterns. The combining step of the divide-and-conquer solutions uses efficient sequential techniques for merging information about subimages. The proposed solutions are implemented on Intel Delta and Intel Paragon machines. We discuss performance results and present scalability analysis using different image and machine sizes.

Scalability of 2-D wavelet transform algorithms: analytical and experimental results on MPPs

This paper studies the scalability of two-dimensional (2-D) discrete wavelet transform (DWT) algorithms on massively parallel processors (MPPs). The principal operation in the 2-D DWT is the filtering operation used to implement the filter banks of the 2-D subband decomposition. This filtering operation can be implemented as a convolution in the time domain or as a multiplication in the frequency domain. We demonstrate that there exist combinations of machine size, image size, and wavelet kernel size for which the time-domain algorithms outperform the frequency domain algorithms and vice-versa. We therefore demonstrate that a hybrid approach that combines time- and frequency-domain approaches can yield linear scalability for a broad range of problem and machine sizes. Furthermore, we show the effect of processor speed versus communication overhead and the use of separable versus nonseparable wavelets on the crossover points between the algorithm approaches.

Parallel scalable libraries and algorithms for computer vision

We describe a project that integrates applications requirements, parallel algorithm design, models of parallel computing, and software tools in order to improve the ability of applications researchers in the fields of computer vision and image processing (CVIP) to realize the performance potential of high performance parallel computers. This objective is achieved by pursuing four directions of research: the development of efficient and practical scalable algorithms for fundamental CVIP problems; the implementation of realistic CVIP scenarios on high performance computers; the development of a scalable, architecture-independent parallel model and a metric for estimating the performance of an algorithm on an existing parallel machine; and the development of prototype software tools. In this paper we describe work in each of these areas.

Evaluating scalability of the 2-D FFT on parallel computers

Parallel computers have demonstrated a remarkable potential for achieving high performance at a reasonable cost for many computer vision and image processing (CVIP) applications. A major obstacle to the use of parallel computers is the lack of a universally accepted metric to study the scalability of parallel algorithms and architectures. The authors apply different scalability measures to various 2-D FFT algorithms and target architectures and compare the expected performance to the measured results. A number of algorithms in computer vision and image processing exhibit regular communication patterns similar to the 2-D FFT. The authors can therefore extrapolate the observations to determine which aspects of these measures are relevant to the scalability analysis of other similar image processing algorithms.

Scalable parallel list ranking of image edges on fine-grained machines

We present analytical and experimental results for fine-grained list ranking algorithms, with the objective of examining how the locality properties of image edge lists can be used to improve the performance of this highly data-dependent operation. Starting with Wyllies (1979) algorithm and Anderson and Millers (1990) randomized algorithm as bases, we use the spatial locality of edge links to derive scalable algorithms designed to exploit the characteristics of image edges. Tested on actual and synthetic edge data, this approach achieves significant speedup on the MasPar MP-1 and MP-2, compared to the standard list ranking algorithms. The modified algorithms exhibit good scalability and are robust across a wide variety of images.<<ETX>>