Indeed the problem is not with star detection, but just the opposite: too many stars
The images can be registered without problems by just increasing the noise scales parameter (star detection section) up to 3, and leaving the rest of parameters with default values, including detection scales = 5.
These images have a significant scale difference of 1:0.67. The current star detector, as noted in the documentation, has some difficulties with this. Summarizing, too many stars are being detected (comparatively) in the large-scale image, which leads to a complex star matching problem with very high uncertainty due to proliferation of small-scale triangles. It seems the changes I have introduced aggravate this problem because the star detector is now more sensitive. By increasing the noise scales parameter, only the brightest stars are detected in a consistent way on both images, which simplifies the problem by removing billions of small-scale triangles.
Thanks for an interesting test case. It is obvious that we need a better star detector to deal with large scale differences.
