Applying incremental learning to parallel image segmentation

Segmenting large or multiple images is time and memory consuming. These issues have been addressed in the past by implementing parallel versions of popular algorithms such as Graph Cuts and Mean Shift. Here, we propose to use an incremental Gaussian Mixture Model (GMM) learning algorithm for parallel image segmentation. We show that our approach allows us to reduce the memory requirements dramatically whilst obtaining high quality of segmentation. We also compare memory, time and quality of the performance of our approach and several other state of the art segmentation algorithms in a rigorous set of experiments, which produce favorable results.