Input
Superpixel Segmentation
Mean RGB
Description :
A superpixel is a local group of pixels sharing similar colors. Segmenting an image in superpixels has many advantages. Instead of working at pixel level, working at superpixel level reduces the number of units to process. Secondly, a superpixel fits to object boundaries, which makes it a great tool for object segmentation.
However, a superpixel algorithm should be fast to be worth considering as preprocessing. Therefore, I implemented a GPU version of SLIC (SLIC Superpixels Compared to State-of-the-art Superpixel Methods, Achantal et al. , 2011), which is able to run at 24 FPS on a 1920x1080 image with a GTX 970M.
Tools :
C++, OpenCV and CUDA.
Source code :
SLIC CUDA is now integrated into the LITIV Computer Vision Framework but you can still find a standalone version here.