Extraction of Tiled Top-down Irregular Pyramids from Large Images

Processing large images is a common issue in the computer vision framework with applications such as satellite or microscopic images. The major problem comes from the size of those images that prevents them from being loaded globally into memory. Moreover, such images contain different information at different levels of resolution. For example, global features, such as the delimitation of a tissue, appear at low resolution whereas finer details, such as cells, can only be distinguished at full resolution. Thus, the objective of this paper is the definition of a suitable hierarchical data structure that would provide full access to all the properties of the image by representing topological information. The idea consists in transposing the notion of tile for top-down topological pyramids to control accurately the amount of memory required by the construction of our model. As a result, this paper defines the topological model of tiled top-down pyramid and proposes a construction scheme that would not depend on the system memory limitations.