Hi Rodd,
These artifacts are being caused by interpolation
aliasing, and are a variant of
moiré pattern. This happens because the images have a considerable amount of noise and the registration transformations have to apply small rotations. See
my article about interpolation algorithms (old document, but the information given is still valid). If you prefer a quick read, you can jump directly to the
Rotation Examples section.
As you can see in the comparison, Lanczos interpolation generates minimal aliasing artifacts compared to the rest of algorithms. That's why it is the default interpolation in StarAlignment. Nearest neighbor interpolation generates no artifacts by preserving the original noise distribution, but it lacks subpixel registration accuracy, which is a serious problem unless you work with oversampled data (and even in such case, subpixel registration is always desirable IMO).
The best ways to prevent (or nullify) these problems are:
- Acquire more signal. With less noise aliasing problems can be much less severe.
- Acquire many images with good dithering. By integrating a large set of dithered images, aliasing artifacts tend to cancel out.
- Use drizzle integration to generate your final integrated image (drizzle x2 if you work with undersampled data; x1 if the data are well sampled or oversampled). Since drizzle does not apply any interpolation, there are absolutely no aliasing problems.
I hope this helps you to understand the problem, and hence to potentially solve it.
