I checked the code, you use only the first part of Xenowhirl's not-completely-documented algorithm (apply scale2x up to 3 times, rotozoom back to original size).
It's the same that I experimented with grafx2 as the interactive brush resizer. I have the same comment as Argyle:
I end up just generating several images within the angle range I'm looking for, blindly hoping that one is a result that will be useful
In my opinion you should think of the following 2 features:
1) Allow more precision than integer degrees. Sometimes 44.9? or 45.1? give much better results than 45?
2) Allow picking the offsets of subpixels : When you perform the rotozoom, instead of looping on 0, 1, etc. up to size-1,you can use 0.5, 1.5 etc. up to size-0.5 This would be an offset of 0.5, meaning you pick the middle of pixels instead of their left border. Since the picture is 8 times bigger than original, you can choose between 8 different offsets, and all will give subtle variations: 0, 0.125, 0.25, 0.375 etc... up to 0.875. Since you can choose different offsets on X and Y, there are 64 possibilities. It's impossible to predict which offsets will give the best visual result.