Tests of the photometric accuracy of image restoration using the maximum entropy algorithm

Abstract

Simulations are described which test the maximum entropy image restoration algorithm as implemented in the MEMESYS3 code (version 2) of S. F. Gull and J. Skilling [Quantified Maximum Entropy, MEMSYS3 Users’ Manual, version 2.0 (1989)]. It is found that at the faintest brightness levels, while the code can recover blurred point sources, they are recovered systematically too faint. The size of the photometric error depends on the brightness of the source (relative to the noise of the background) and the crowding of the field. The error increases substantially as the crowding increases. At present, the optimum technique to apply to blurred images of crowded fields where most of the sources are point sources seems to be to use a restored image to generate the list of objects, then feed this into a standard point-spread function-fitting code (such as DAOPHOT) and use this on the original blurred frame. In that manner, the most crowded fields can be analyzed without losing photometric accuracy. Additional simulations were carried out for images with the actual point spread function of the Wide Field Camera of the Hubble Space Telescope. For single isolated sources seen against a background characterized by Gaussian noise, the detection limit is degraded near its faint limit by a factor of about 4 compared to that expected. But reliable photometry cannot be obtained for sources at the detection limit in either the simulated frames, or those frames passed through the MEMSYS3 image restoration code. If one requires photometry accurate to 10%, the performance of the as-built HST plus Wide Field Camera is degraded near its faint limit by a factor of between 10 and 15 compared to that expected, even for isolated point sources.