Image denoising using CUFFT

A parallel implementation for image denoising on a Nvidia GPU using Cuda and the cuFFT Library The sofware:

• Automatically selects the most powerful GPU (in case of a multi-GPU system)
• Executes denoising
• Saves output as text file and image

Authors

• Vincenzo Santopietro <[email protected]>
• Francesco Battistone <[email protected]>

Installation

The software requires OpenCV v2.4.x to run.

Install the dependencies and compile the Makefile.

$ cd image-denoising-using-cufft
$ make

Demo

The software needs some parameters:

• Number of threads (per block, for each direction)
• Path to the output '.txt' file. If the file does not exist, it's created.
• Path to the input image
• Path to the '.txt' containing the kernel. [NB kernel's size is fixed to 5x5]

You can run the software by typing

$ ./convolution <num_threads> <output_file_path> <image_path> <kernel_path>

Plugins

Plugin	README
OpenCV	[https://github.com/opencv/opencv/blob/2.4/README.md] cv
cuFFT	[http://docs.nvidia.com/cuda/cufft/#axzz4f6kyGEZu] df1

Version

This is still a beta version, developed for academic purposes.

Development

Want to contribute? Great!

Todos

• Add an input parser

License

GNUv3.0
If you use this software, please cite: http://www.sciencedirect.com/science/article/pii/S1877050917317209

Credits

Image: https://en.wikipedia.org/wiki/Gaussian_noise

Free Software, Hell Yeah!