Home > Published Issues > 2015 > Volume 3, No. 2, December 2015 >

GPU Based Fast Non Local Means Algorithm

Daniel Sanju Antony and G. N. Rathna
Electrical Engineering, Indian Institute of Science (IISc), Bangalore, India

Abstract—Non Local Means (NLM) Algorithm proposed by Buades et al., gave remarkable denoising results at expense of computational cost. Darbon et al. used the separable property of the algorithm to create a faster implementation. In this paper, parallelization of this modified Non Local Means denoising algorithm using Heterogeneous computing platforms like Central Processing Unit (CPU) and Graphical Processing Unit (GPU) is developed. The algorithm is implemented on GPU with the help of OpenCL API. Experimental results show that the GPU based implementation is about 25 times faster than the CPU based implementation of Buades et al. algorithm and around 85 times faster than Darbon et al. implementation.

Index Terms—heterogeneous computing, denoising, OpenCL, CPU, GPU

Cite: Daniel Sanju Antony and G. N. Rathna, "GPU Based Fast Non Local Means Algorithm," Journal of Image and Graphics, Vol. 3, No. 2, pp. 122-125, December 2015. doi: 10.18178/joig.3.2.122-125

PREVIOUS PAPER
Reconstruction of Premature Atrial Contraction and Premature Ventricular Contraction on ECG Traces by Applying PLA as Segmentation Process
NEXT PAPER
Augmented Reality and Situation Awareness Applications for Military Computing