As an example of typical results, Fig. 1 shows a simulated HST Wide Field and Planetary Camera (WF/PC I) image of a distant cluster of galaxies. Its construction was described in Freudling and Caulet (1993). The upper left is the ``ground truth'' image, the upper right the image degraded through convolving with a spherically aberrated PSF and adding noise, the lower left is a restoration with noise removal, and the lower right is a Richardson-Lucy restoration. These restorations used, respectively, 28 and 40 unaccelerated iterations using a Gaussian noise model. Various tests were carried out on the object detection potential of the two restored images, with a conclusion pointing narrowly in favor of the noise-suppressed restoration. Further details are to be found in Starck and Murtagh (1993).
Another area where the strategy described in this paper works well is when the image background varies. Richardson-Lucy, without intervention in wavelet space, must first model the background and remove it. It appears that the wavelet-based approach handles such a situation well, and without such removal of the background, due to the fact that it detects significant structures in differenced (smoothed) images. An example of this may be found in Starck and Murtagh (1993).