Image-Processor

The project contains sequential and different parallel implementations of image processing algorithm.
The project was created for the needs of master's thesis of Computer Science studies.
The aim of the project was to compare performance of sequential and parallel implementations, depending on threads and processes number used for computing, with the use of different approaches and technologies.

Functionality

The selected image is processed by applying a chosen filter e.g. blur/sharpen/edge detection/emboss/outline or any other by providing appropriate kernel matrix.
The input image is loaded from images directory by entering image name in code.
Modified image is saved in the same directory.

Filter can be chosen in main function by modifying:

auto filter = Filter::blurKernel();

A few kernels are available to choose from but new can be easily added. The result may be poorly visibly on high resolution images as the project was created to analyze performance, not to focus on functional image processing aspects.

Technologies

• C++
• OpenMP
• OpenMPI
• CUDA
  • Thrust

Implementations

The project contains a few image processing implementations:

• Sequential implementation - using pure C++
• Multithreaded implementation with the use of OpenMP
• Parallel implementation with the use of OpenMPI
• Massive parallel implementation with use of CUDA

Installation

Dependencies

• Visual Studio
• OpenMP support enabled in Visual Studio (for ImageProcessor-OpenMP project)
• OpenMPI implementation installed e.g. Microsoft MPI
• CUDA installed (+ CUDA capable graphic card)
• Thrust

Repository

$ git clone https://github.com/marmal95/image-processor.git

Build

The Visual Studio solution contains four projects inside - responding four implementations mentioned in Implementations section.
Build whole solution by choosing:

Build > Build Solution

from top menu, or right-click specific project in Solution Explorer and choose:

Build

Run

Right-click on chosen project in Solution Explorer view and click Set as Startup Project.
Click F5 or choose Debug >> Start Debugging from top menu.

Customization

OpenMP

Preferred number of threads in OpenMP implementation used for computing may be changed with function call:

omp_set_num_threads(NUM_THREADS)

which is called at the beginning of main() function.

OpenMPI

Preferred number of processes is passed as parameter to mpiexec command.
The value may be changed in Visual Studio in: Project > Properties > Configuration Properties > Debugging.

Command             mpiexec.exe
Command Arguments   -n 4 "$(TargetPath)"

CUDA

Preferred size of grid used for computing is specified inside CudaAlgorithm.cu file.

dim3 threadsPerBlock(16, 16);
dim3 numBlocks(..., ...); // currently calculated based on image size
applyFilterOnCuda<<<numBlocks, threadsPerBlock>>>(...);

Inside that call the number of blocks and number of threads per each block is specified.