high performance computing on graphics processing units: hgpu.org

This thesis investigates the suitability of OpenCL for acceleration of Image analysis operations from a developers perspective. To achieve this four representative problems: Morphological operations, Convolution, Watershedding and Markov random field-based texture segmentation are evaluated. The selected problems offers different implementation issues in terms of locality of the operations and load versus computation. The thesis covers recent research into these topics as well as offers proof-of-concept implementations and benchmarks of some algorithms. The thesis also covers the process of working with OpenCL and the reliability and tool support for OpenCL on different platforms. Finally, the results of the proof-of-concept implementations is put in relation to the development process in a effort versus performance comparison and the suitability of using OpenCL for a general Image analysis framework and application integration is discussed.