Chapter 4 – Neighbor Embedding

This chapter explores the internal learning option of decomposing each image patch as a linear combination of example patches, instead of trying to generalize from a single example as in high-frequency transfer. The neighbor embedding framework is presented and its application to super resolution is discussed. This chapter focuses first on internal learning (examples are extracted from transformations of the input image), and moves toward external learning, which enables single-step super resolution. Finally, a discussion on the practical shortcomings of the application of neighbor embedding for external learning leads to the introduction of more advanced external learning techniques presented in the next group of chapters.