samminweng / image-processing

Image Processing using CUDA (C++ & Python)

CUDA Image Processing

• CUDA C++ Image Processing Project for High Performance Computing (cudacpp)
• PyCUDA Image Processing Project for High Performance Computing (pycuda)

Image filters implemented

Grayscale

• Formula:

gray = (Max(R, G, B) + Min(R, G, B)) / 2

Gaussian Blur

• Gaussian Function:

G(x, y) = (1 / 2 * PI * sigma²) * exp ^{(-(x² + y²))/ 2 * sigma²}

Notes

• Different branches for CUDA C++ & PyCUDA Implementations (cudacpp & pycuda)
• Implementation using C++11 & CUDA 8.0
• Implementation using PyCUDA & CUDA 8.0
• Windows 10 | Visual Studio 2015
• Used Photoshop RAW image (*.raw) for reading the RGB channels
• Host code implemented for CUDA C++ only
• Device code implemented for PyCUDA & CUDA C++
• Contains default stream (stream 0), non-default stream and multi-streams codes for Gaussian Blur for CUDA C++ implementation
• Multi-streams implementation not working fully (image contains black lines for number of stream - 1)
• Depth-First multi-streams implemented, Breadth-First faster.
• Same kernel function used for PyCUDA & CUDA C++

Reference

• Gaussian Blur
• Grayscale Algorithms