Efficient morphological tools for astronomical image processing

Nowadays, many applications rely on a huge quantity of images at high resolution and with high quantity of information per pixel, due either to the technological improvements of the instruments or to the type of measurement observed. This thesis is focused on exploring and developing tools and new methods in the framework of Mathematical Morphology, suitable for automatic image processing of astronomical datasets.

Identifying and classifying astronomical objects is a challenging task: many of the structures that astronomers are interested into to understand the evolution of galaxies and stars are faint and near to the noise level. A new method to identify astronomical objects is proposed, based on the expected statistical behaviour of the noise across the image structures. The method works by parsing a hierarchical representation of the image data, called max-tree: it allows for image filtering and object identification in an efficient way. A novel parallel algorithm to build max-tree of 2D images and 3D volumes with very high-dynamic-range values is proposed. Besides astronomy, other application fields, like medical and remote sensing image processing, would benefit from that. Effort is put also in classifying the objects found. A parallel method is developed to compute efficiently the pattern spectra. Those are matrices whose values put in relation area and shape information of selected structures in the image. Such matrices are used both to classify different types of galaxies and to identify building footprints in satellite images.