Hi all,
As anticipated in previous forum posts, the next version of the StarAlignment process includes two important new features:
Automatic intersection computation. When this option is enabled, StarAlignment automatically computes the intersection region between target and reference images, using a special FFT-based algorithm. Once the intersection has been found, the star matching process is restricted to the intersection region. This greatly improves the performance of the RANSAC routine by minimizing the fraction of outliers (false matches) in the initial set of putative star pair matches. In essence, this works like the manual procedure based on previews to define overlapped regions, which has been available since PI 1.5, but in a completely automatic (and much more accurate) way. The preview-based procedure is still available but it is only required in extremely difficult cases, when automatic intersection computation fails due to very little overlapping between mosaic frames.
Automatic mosaic frame adaptation. With this option selected, StarAlignment fits a linear function to adapt background and signal levels of each target image with respect to the reference image. The linear fit is of the general form:
y =
a +
bxwhere
a and
b are the two fitting parameters. The linear function is computed from all overlapping pixels between the target and reference images. The
a parameter acts like a pedestal to compensate for different background illumination, and the
b parameter works like a scaling function to equalize signal levels. Frame adaptation is extremely useful to create seamless mosaics. Bear in mind, however, that the linear fit assumes that (i) both images are strictly linear and (ii) both images have flat illumination profiles, i.e. that both images have been calibrated with accurate flat-field frames.
In a recent video tutorial, we demonstrated how mosaic frame adaptation can be implemented as an iterative procedure with PixelMath using mosaic frame masks. The automatic frame adaptation algorithm performs the same task but is much more accurate and completely automatic. Below you can see a few examples that show the kind of result that can be easily achieved when the images have been properly calibrated.
<
Click here to enlarge>
The screenshot above shows a two-frame mosaic of M31 acquired by Jordi Gallego, who has kindly allowed me to use his raw images for testing purposes. Note that thanks to the automatic intersection computation feature, this mosaic can be generated in a completely automatic way, without the help of previews, in spite of the fact that the overlapping area between the two mosaic frames is minimal (a 4% approx. of overlapped surface). The parameters that control the two new StarAlignment features have been marked with green ellipses. Intersection computation is enabled for mosaics by default, so normally there is no need to change its default setting. Frame adaptation must be explicitly enabled.
Below you can see, to the left, the result with mosaic frame adaptation enabled. To the right, the same mosaic with frame adaptation disabled.
<
Click here to enlarge>
Below is another example with Nikolay Volkov's four-frame mosaic of the California nebula. This is the same image used in the video tutorial noted above. Again, the results with frame adaptation enabled and disabled can be seen to the left and right of the screenshot, respectively.
<
Click here to enlarge>
More news on 1.6 very soon. Stay tuned!
