Image colourisation using graph-based semi-supervised learning

A novel colourisation algorithm using graph-based semi-supervised learning (SSL) is presented. We show that the assumption of the colourisation problem is consistent with the fundamental of graph-based SSL methods. Satisfactory results are obtained in the experiments that validate the proposed algorithm. To reduce the time and memory requirements when dealing with large scale images, we further propose a two-stage speedup approach. Comparative results show that the computation complexity is dramatically reduced.