Image Denoising

Implementing a parallel algorithm for image denoising with the Ising model using Markow Chain Monte Carlo. Note that by images, black-white images (Images that consist of black and white pixels only) will be meant. And these black and white pixels will also be referred as -1 and 1, which is the representation of these images with text, provided in the project. We implemented 3 versions of this algorithm: sequential version, Pthreads version and MPI version.

Sequential version

How to compile

$ gcc denoiser_sequential.c -o denoiser -lm

How to run

$ ./denoiser <input_file> <output_file> <beta> <pi>

Pthreads version

How to compile

$ gcc denoiser_PTHREADS.c -o denoiser -lm -lpthread

How to run

$ ./denoiser <input_file> <output_file> <beta> <pi>

MPI version

How to compile

$ mpicc denoiser.c -o denoiser -lm

How to run

Run the program using command (by default it will be run in grid mode):

$ mpiexec -np <nof_processors> ./denoiser <input_file> <output_file> <beta> <pi>

Optionally, it can run by the following command to run in row mode:

$ mpiexec -np <nof_processors> ./denoiser <input_file> <output_file> <beta> <pi> row

where:

• <nof_processors> Number of processors must satisfy a special condition, let’s say nof proccessors is n, then, in row mode the width and height of the image must be divisible by (n-1), in grid mode the width and height of the image must be divisible by sqrt(n–1) and sqrt(n-1) must be an integer.
• <input_file> A text file which includes a n * n black-white image described with 1 and -1 values Each row must be terminated by a line ending and each column must be separated by a space.
• <output_file> A file name to output denoised image, again as a n * n black-white image described with 1 and -1 values where each row is terminated by a line ending and each column is separated by a space.
• The beta value to be used as the beta value in the Metropolis-Hastings Markov Chain Monte Carlo functions.
• The pi value to be used as the pi value in the Metropolis-Hastings Markov Chain Monte Carlo functions.
• row That’s an optional parameter to run the program in row mode, by default the program will be run in the grid mode. (An exact string “row” must be provided as the argument)

Program outputs the denoised version of the initial image to the given <output_file> path. It is the text version of a black-white image represented as a grid of 1 and -1 values.

Example of denoising image

Since the program works with text files containing -1 and 1, we need to convert the png image to text, for this we use a simple script python scripts/image-to-text.py. Another script is scripts/make_noise.py, this will add noise to our 'image' (text) and will be the input of parallel program. Parallel program will produce an text file with denoised image and with scripts/text_to_image.py script will get a denoised image.

Original	Noisy	Denoised

Authors

• Raffaele Franco - Initial work - raffaelefranco
• Lorenzo Gagliani - Initial work - lorenzosax