Star Detection

The first criterion used to evaluate the restoration techniques is the ability to detect stars in the restored image without ``detecting'' noise and restoration artifacts. We used a simple detection algorithm which finds the local maximum in the restored image above a user-supplied threshold. Fig. 4 shows how the number of stars detected (solid line) increases as the threshold decreases. At the same time, the number of false detections (dashed line) also increases. As the threshold is decreased below 100, more than half of the additional detections are spurious.

Table 1 gives the results for various percentages of false detections. The first two methods (both linear) show very poor results. Numerous artifacts around bright stars make star detection very difficult in the linear restorations. The next three methods (all non-linear) give significantly better results. The R-L/Snyder method gives slightly better results than the R-L and MEM methods. Some improvement in star detection is seen in the hybrid method when is large (0.1). However, as decreases, artifacts begin to show up in the solution.