OpenCL Resizer

Goal

Goal of this project is to explore performance of OpenCL and its applications on image processing.

What it does

It's a high performance image resizing utility. It supports JPEG images and 3 sampling methods:

• nearest neighbor
• linear interpolation
• cubic interpolation

How fast is it

For high resolution images (2k+) it is faster than OpenCV. It's not particularly fast for low resolution images due to overhead of data transfer between CPU and GPU.

How does it work

It resizes images on GPU (when found) rather than CPU. Detailed explanation can be found on my blog http://vaclavsamec.com/blog_resizer.

How to use it

Call resizer with these parameters:

Parameter	Description	Default
-input	input directory of JPEG images	required
-output	output directory of resized images	required
-quality	quality of output JPEG image in range <1, 100>	95
-algo	sampling algorithm, use one of these [nn, linear cubic]	cubic
-ratio	ratio of resizing	1.0

Example

resizer -output ./testImages/output -input ./testImages/input -quality 95 -algo cubic -ratio 0.5

How to build it

You need C++ compiler and following libraries:

• OpenCL
• Libjpeg Turbo
• CMake

See CMakeList.txt for further adjustments of libraries path.