Fast Normalized Cut for Image Segmentation on the GPU

Recent advances in the speed and programmability of graphics hardware permit the GPU to grow as a powerful vector coprocessor to the CPU. In this work, the emphasis will be to implement fast Matrix-Vector operations to improve techniques for eigenvalue decomposition. We introduce a framework for the implementation of this mathematical operation, thus providing the building blocks for the design of more complex numerical algorithms. In particular, we propose a stream model for performing a faster Normalized Cut algorithm on image segmentation problems. By implementing the calculations on the GPU, we hope to achieve a performance gain, or at minimum free up CPU resources. With the goal to experiment the algorithm on a real sequence of images and report the results of our application.