Did some tests with various cover images, and the patterns in my previous screenshots seem to be more of an exception than a norm.
Those patterns must be some sort of artifact resulting from high magnification of a compressed digital image. Usually they can be observed from edge areas where different, contrasting colors meet. That's why they are more pronounced in multicolored images like "Tango in the Night".
If you later implement controls over blur intensity and magnification power, I could do more investigation. But I'm fine if you replace it with a new algorithm for TM since only a small fraction of covers get affected by artifacts with greater improvement in general.