Home Page Wed Oct 15 02:17:40 JST 1997
Next: Conclusion and future work Up: Technical Details of PIXY Previous: Star detection from an |
Matching between an image and a catalogMatching means to obtain a map function with the two lists of coordinates and brightness: detected stars list from an image and extracted star data list from a catalog. If which data is the counterpart of any star in the image is clear, the map function can be calculated with any two pairs.
But actually the two lists are not related to each other. Therefore the system calculates many map functions with many probable counterparts and tries to find out true map function among them. In this way, it expects that the value of map function calculated with wrong counterparts hardly overlaps and the value of true map function is obtained most frequently. Definitely, the process of matching is as follows.
First of all, I made an experiment with a test data. The data is a 600 400 virtual night sky containing 200 stars in it. The difference of the brightest and the faintest is 10 steps. Then the system makes a copy of a partial rectangular region. It becomes an artificial photo image. The width and height of the image is about of the sky. Because there should be some errors in actual photographs, the system changes the position and brightness slightly when copying: about pixels in position, about steps in brightness. In addition, 1 - 4 new objects are also appended. Then it tried to calculate the map function between the artificial photo image and the sky. In this experiment, the approximate position input beforehand should be in the area of the image, and the approximate width of the image should differ less than 20% errors from the true value. As a result of trials so that the system could tell where the image should be in the virtual sky correctly, the process mentioned before is implemented as follows.
Now that the system deals with real images, it has to convert the brightness of detected stars from an image to magnitude to compare with stars in catalogs in matching. The process is as follows.
By the way, real images contain much more stars than the test data and the implementation determined with the test data was found to yield many wrong map functions. So I improves the implementations again as follows.
Here I show you the results of experiments with real images. I use the Tycho Catalog as a real star catalog. The image upper left is the original one. The center in the left side shows the detected star list from the image. The large image in the right is the chart around the approximate position, whose data are from the Tycho Catalog. The bottom one in the left side is the result, the system re-mapped the extracted stars with the obtained map function and made the chart of the same area as the original. When the result chart is quite same as the original, it means the system could obtain the true map function.
In these experiments, true map function was obtained every time independent on the number of stars. However, there are two big problems actually and they obstructs to open the system in the public.
By the way, I have also tried the way to select four stars and check if a pair of quadrangles s similar, but only to fail. The system turned to take enormous time because the exponent increased. In addition, the parameters could not converge on the true value. I have also tried the reversed process. That is, the system actually mapped all detected stars with each candidate of map function, definitely the center of each block in the parameter space, counted how many stars mapped exactly on stars in a catalog, and found the parameter with the most exactly mapped pairs, but only to fail either. It seemed to take extreme time to finish and I interrupted it while running.
|
Next: Conclusion and future work Up: Technical Details of PIXY Previous: Star detection from an |