My understanding of copyright law [1] is that the maze itself, not the maze-generation process, is the issue here. The author accrues a copyright over the maze by virtue of having published it. Kraft's republication of the maze is potential infringement of the copyright over the maze. The algorithm is almost irrelevant in that determination. The copyright is over the work itself, not the process of creating the work. [2]

The test for determining infringement is, more or less, the classic smell test: does it look like a duck, walk like a duck, and quack like a duck? Would a reasonable individual conclude, from visual inspection, that Kraft's maze is strikingly similar to the original maze? Even if there are some small changes here and there, it could still be found to be sufficiently similar.

For instance, if I tried to put out a version of a popular song that was ever so slightly different from that song, I'd be infringing if a reasonable listener found my version sufficiently close to identical. If I put out a book called "Jimmy Potter and the Sorcerer's Stone," and my work was more or less identical to "Harry Potter," with some character names and minor sequences changed, I'd probably be found to be infringing upon the original work.

[1] IANAL, but I've dealt with copyright issues, to a certain tangential extent, in a professional capacity.

[2] Note that we're talking about copyrights here, not patents, which are different animals altogether.

Cynically, I would say that it depends on whether or not you can get your case to the Supreme Court.

Normally, though, algorithms are not copyrightable. Though Books about algorithms are.

In this case, the guy published a book containing a maze, so reproduction of the maze should fall under copyright.