User transparency: a fully sequential programming model for efficient data parallel image processing

Abstract

SUMMARYAlthough many image processing applications are ideally suited for parallel implementation, most researchers in imaging do not benefit from high-performance computing on a daily basis. Essentially, this is due to the fact that no parallelization tools exist that truly match the image processing researcher's frame of reference. As it is unrealistic to expect imaging researchers to become experts in parallel computing, tools must be provided to allow them to develop high-performance applications in a highly familiar manner. In an attempt to provide such a tool, we have designed a software architecture that allows transparent (i.e. sequential) implementation of data parallel imaging applications for execution on homogeneous distributed memory MIMD-style multicomputers. This paper presents an extensive overview of the design rationale behind the software architecture, and gives an assessment of the architecture's effectiveness in providing significant performance gains. In particular, we describe the implementation and automatic parallelization of three well-known example applications that contain many fundamental imaging operations: (1) template matching; (2) multi-baseline stereo vision; and (3) line detection. Based on experimental results we conclude that our software architecture constitutes a powerful and user-friendly tool for obtaining high performance in many important image processing research areas.

High-performance SIMT code generation in an active visual effects library

“…In [14] the authors describe a library for shared and distributed memory CPU parallelisation of computer vision algorithms expressed in image algebra, which encompasses the functionality of our point and filter indexers. Complete effects are similarly built from primitive operations into DAGs from straight-line code.…”

Towards fully user transparent task and data parallel image processing

“…The task dependency graph is automatically generated at runtime which is used for interoperation optimization. This overcomes the main limitation of a librarybased approach [11]. In our approach, image operations are specified at a higher level than for example in [7].…”

“…The key extensions to the state-of-the art [6,13,7,11] of user transparent parallel image processing are the following:…”

Streaming Applications

“…It is based on line detection applications found in [107], which use second order Gaussian derivative convolution kernels. In our application, we use first order Gaussian derivative convolution kernels.…”