...$(\sqrt{170}-13)/13 = 0.3\%$.^2.1

In [37], Danielsson overlooks the existence of this error and announces the maximal relative error to be $0.15\%$

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

... image^2.2

For arbitrary dimensions, it is actually both easier and more efficient to extend Ragnelmam's ordered propagation algorithm [127] than to use his corner EDT [128]

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

... images^3.1

This remark may be hardware dependent. In particular, PMON should out-perform PMN in all cases on a SUN Sparc workstation, where the relative cost of the floating point operations is lower.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

....a^4.1

The illustrations found in this section are taken from [45]. This online anatomy course has also largely inspired the text of the section

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

... checked^6.1

Actually, to avoid unnecessary computations, it is only used for pixel p in the border of the object O, i.e. such that $p \in O$ and $\exists q = p+n \in O'$ with a direct neighbor n

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

... Practically^6.2

This limit is of course dependent on the content of the image. In particular, it would be possible to find pathological images - such as lines oriented at 60^O in all slices - where the limit is much lower. For other images - such as a single object pixel in the middle - Saito's algorithm is always the fastest. Nevertheless, the ``sphere'' image gives us a good indication of where the limit will be in most practical cases.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

... PSN^6.3

At figure , PSN's computational cost increases for images sizes over 700. This contradicts the theory that says that PSN has a fixed cost per pixel. We postulate this experimental increase is linked to the size of the dynamic data structure that holds the pixels in the propagation front. For small images, the buckets' structure is small enough to hold entirely within the cache memory. For large images, it has to be stored in the main RAM, which slows down the whole process.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.