Overwriting the original PDF and outputting another PDF is definately not what it is supposed to be doing. I've found and fixed the problem. And I think I've found how to fix the .net error, but Dave will let me know regarding that.
I've also deleted all the captions referencing the evil AutoImager.
PDFrasterfarian 2.2.1
"Sweet! This is nice, I'll have to give it a whirl and compare it with previous runs in your earlier version. Must've been a bugger to write your own filter to apply (a friend of mine did one with a stego-detection program he wrote, but it was done it Java... I think). Thanks Alex!"
Actually, the dilate filter turned out mind-numblingly easy. You just read in nine pixels and output the biggest of the bunch. I implemented it as a native C++/CLI library that I call from a .NET executable (in which i read the png into memory and get a pointer to the raw data). Implementing an image filter in Java must have given some horrible performance.
But anyway, please compare the new post-processing. It's easier to read but maybe not as nice to look at as the older fuzzier version. There's also some concern that the mode that i now run ghostscript in will render some fonts incorrectly. I think I'll restore the old processing as an option under Advance mode. If you'd like, you can also try tweaking kernel-processing.cmd/kernel-preview.cmd to see if maybe you can arrive at better settings.
Some notable lines:
line 73/74 which controls ghostscript (the new param -dGridFitTT=3 may give problems. Old behavior was =2)
line 82/83 which runs the contrast-enhancing filter. (new param is -unsharp 2x0.8+2.0, old is -unsharp 2x1+1)