In a scale-space theory, an image is embedded into a continuous family of gradually smoother version of it. The original image corresponds to the finest scale t = 0, and increasing the scale should simplify the image towards coarser scales without creating spurious structures. Since a scale space introduces a hierarchy of the image features, it constitutes an important step from pixel-related image description to a semantical image description.
This project aims at exploring deep scales for two main purposes: one is to study the possibility of moving back from a coarse scale to reconstruct an underlying image structure at a fine scale; the other is to examine a coarse scale image, in a scale-space framework, to figure out which pixels contribute in forming a specific pixel at a coarse scale.