Parallelizing A Content-Aware Image Resizing with OpenSHMEM

Abstract

ABSTRACT Cropping and Scaling are the most common image resizing techniques but neither of these considers the content of the image. Seam carving is a content-aware image resizing technique that removes or duplicates the least visible pixels from an image, thereby making it more effective than Cropping and Scaling. For large batches of images, it may be unrealistic to do Seam carving on one processing element due to memory constraints. To solve this problem, a parallel approach to image resizing that helps to ease data transfer between processing elements needs to be considered.

Partitioned Global Address Space (PGAS) programming models have been attracting attention as a parallel computing model and it is often used to implement one-sided Remote Memory Access (RMA) from multi-host systems, such as computer clusters. OpenSHMEM is a distributed, PGAS programming model that has light-weight semantics and high performance RMA and atomic memory operations.

In this thesis, we parallelize Seam carving using Pthreads, OpenSHMEM and MPI. We evaluate the relative performance gained with multiple threads and processing elements (PEs).