A Load-Balanced Parallel Algorithm for 2D Image Warping

摘要

2D image warping is a computation-intensive task and plays an important role in many fields. All existing parallel image warping algorithms are only suitable for limited geometric transformations. This paper proposes a distributed parallel image warping algorithm PIWA-LIC, which partitions the valid area of the output image in a balanced way, and each computing node does resampling for one sub output image. To guarantee data locality, a line segment approximation (LSA) method is exploited to get the corresponding input area for each sub output image. For any 2D geometric transformations with one-to-one mapping, PIWA-LIC achieves high performance because of good load balance and data locality.