Parallel PatchMatch using Open-MP Extension

Image editing is an important part of the computational photography. The growing number of digital images in recent years has increased the need to develop powerful editing tools. Every implemented algorithm should be fast enough for users to see the results instantly (real-time manipulation) and control the whole process on the fly. PatchMatch is the state -of-the-art algorithm used in some modern editing tools which are developed to provide an interactive environment for professional image editors. In the classic PatchMatch algorithm, as the images grow in size, the processing time increases drastically making it inefficient to be embedded in a user-interactive tool. One way to overcome this problem is running parallel algorithms in smaller computers which have been made more accessible through the development of multiprocessor computers in the fast-growing personal computer industry. In this paper, we propose a parallel version of PatchMatch algorithm using Open -MP extension. In contrast to the previous parallel versions which use graphical processing unit (GPU), our proposed version only utilizes the existing processing unit and does not require extra hardware resources. Preliminary results show the efficiency of the proposed algorithm in comparison with the sequential implementation.

کلیدواژه ها:

Parallel Computing ، Open-MP ، PatchMatch ، Randomized Nearest Neighbor