Home Page       Tue Sep 2 14:15:44 JST 1997

Experimental result

Finally here shows an experimental result of the prototype system, with a picture of Nova Cas 1995 (Figure 2).

  
Figure 2: Image of Nova Cas 1995

The detected stars from the image are shown as Figure 3. 138 stars are detected. Not only true stars, some flaws of the film and a trianluar mark pointing to the nova are also recognized as stars.

  
Figure 3: Detected stars

It utilizes the Tycho catalog. In this case, I know I took a picture of Nova Cas 1995, so I input the R.A. and Decl. of the nova as the approximate direction. The approximate width is about 1 deg by experience. So the system extracts stars from deg area in the neiborhood (Figure 4). The Tycho catalog contains up to 11 mag stars substantially. But there are some 12 mag stars in the picture, so the density of catalog is less than that of the picture. Anyway, 243 stars are extracted from the catalog. It means, the quarter of 243, about 61 stars should be in the image. Therefore the system picks up 61 stars from the detected list in order of brightness and allots the magnitudes as the average of four stars in the catalog sorted in magnitude. Then it matches the 61 detected stars and 243 extracted stars.

  
Figure 4: Stars in Tycho catalog

As a result of the matching, the following map function is obtained.

k is a rate to magnify, is an angle to rotate and , are offsets. The approximate width was 1 deg and the magnifying rate is 1.15, so the true width is 1.15 deg. The angle 325 deg means the direction 325 deg is north in the image. In other words, after rotating the image 325 deg clockwise, the north becomes up. Then the sytem calculates the inverse of the map function, applys it to the extracted stars from the catalog and creates the chart of the same area as the image (Figure 5). It shows each star in the image corresponds to that in the catalog properly. You can find out easily the nova and the triangular mark are not in the chart. Now that the map function is obtained, the true magnitude of each star in the image is determined.

  
Figure 5: Mapped stars in catalog (same area as the original image)

After the map function is obtained, the system can output the R.A. and Decl. of the position clicked in the image, and show the input R.A. and Decl. is where in the image.

This experiment produced a proper result. However there are some more problems and it takes some more time to publish this system on WWW. The big problems to be absolutely solved are these two things.

The first problem is that matching takes so much time. This experiment took about three hours with PentiumPro 180MHz PC. Even with the approximate angle to rotate, which reduces the search space, it took several dozens of minutes. However it is mainly because the system is implemented in Java language. Before that, I made a experimental program in C++ language. Then I could get result within three minutes, though the target was artificial data, not a real image, and the number of stars were smaller, several dozens of stars.

The another big problem is that simetimes the map function is not determined properly. In this experiment, several candidates of map function such as are listed up. Finally the system adoped the most overlapping (11 overlaps) pair of parameters and it was true. But one of the other candidates has 10 overlaps and there are three candidates with 8 overlaps. So the proper result seems just barely obtained. However, when the approximate angle to rotate is input, within about 30 deg error, the system can obtain the true function.