Home Page       Wed Oct 15 02:17:40 JST 1997

Conclusion and future work

In this article, I described the technical details of the PIXY system, now being implemented, especially the algorithm and technical details of following two points: star detection and matching between an image and a catalog. And I showed some results of experiments with real images.

Many new objects search projects are running at large observatories all over. But they are independent each other and using their own telescopes. Because they yield many images with the same instrument, the process to discover new objects is in image processing, just subtract one from another image of the same area and detects the difference. On the other hand, the MISAO project aims to make use of various images which multitudinous people took for multitudinous purposes. Because the images to be compared are very different each other, the image subtraction method is not suitable. So the system adopts the way that it detects all stars from an image at first, then matches the two sets of dots, detected from an image and extracted from a catalog, geometrically, and regards those dots who has no counterparts as new objects.

As I mentioned in the section of star detection, the current system cannot detect faint objects. Though it cannot be a problem for determination of map function, it can be fatal from the viewpoint of automated new objects discovery, because new objects are generally much fainter. However, the best way to detect much fainter objects really is image processing, to subtract an image from another taken at the same specific observatory under the same condition [3]. The target of the MISAO project is the images taken by various people, not only professionals of sky survey. In this case, the main problem should be the way to reject the influence of such large objects as Moon, bright comets, nebulae and clusters, etc., because people often take images of those remarkable targets. I am looking up how the software SExtractor [4] has solved this problem.

I also listed up two big problems: to take extremely much time and that the system cannot tell one map function as a true one. However, it should be solved by writing the software in Java's native method because the test program in C++ ran very fast. It also solves the latter problem. If the system becomes to be able to take some more time, it can map the detected stars actually with some candidates of map function obtained after the current process and determine which is best.