1996A&AS..120C.683D

We describe a new technique, pixon-based deconvolution, for the reconstruction of images and spectra from low signal-to-noise data. In ``traditional'' techniques, such as Maximum Likelihood (ML) and Maximum Entropy (ME), the model parameters (e.g., pixel size) are considered ``nuisance parameters'', and generally held fixed in the course of deconvolution. Pixon techniques allow the model parameters to change according to the information contained in the data. The pixon model admits only that level of detail which is statistically justified by the data, thus greatly reducing the production of spurious sources and signal correlated residuals, common problems in ML and ME reconstructions. As a result, gains in sensitivity and spatial/energy resolution may be realized. Sample applications to data from OSSE and COMPTEL will be presented.