deblending

Deblending is the process of separating overlapping sources in an image or dataset into distinct components. It is essential in fields where multiple signals share similar spatial or spectral features, such as astronomy, microscopy, and remote sensing.

In astronomy, deblending addresses crowded fields where stars or galaxies overlap due to seeing or instrument resolution. Deblending algorithms attempt to assign pixels to individual sources, estimate their fluxes and shapes, and create separate catalog entries. Popular pipelines use deblending as part of source extraction, with hierarchical subdivisions of blended groups into sub-sources, guided by brightness contrasts and the point spread function.

Methods include PSF-fitting where models of each source convolved with the PSF are fit jointly; multi-object fitting; watershed or seed-based clustering; and probabilistic approaches where each pixel has a probability of belonging to each source. Non-parametric methods and machine learning have been explored.

Challenges include degeneracy when sources are very close, noisy data, variations in PSF, and complex morphologies. Incorrect deblending can bias photometry and object catalogs.

In other domains, deblending applies to separating spectral lines in spectroscopy, separating overlapping footprints in microscopy, or video segmentation where moving objects occlude each other. Evaluation relies on simulated data and ground-truth comparisons.