An isomorphism for digital images (Q1072716)

As the title already suggests, this paper is on the 'equivalence' of digital images, considering this topic from an abstract mathematical standpoint. It proposes the definition of the isomorphism between digital images as a specific relation. In words, this relation states that both can be derived from a continuous pre-image in which the pixels are continuous squares. An isomorphism in such a 'block-image' is shown to induce a digital isomorphism on the derived digital images. It is clearly shown how the different digital connectivities are in this way a direct consequence of the topology of closed and open sets in the continuous pre-image. It should be noted that this relating to a continuous 'block-image' is not the same as the approach of the Mathematical Morphology of Serra (a reference to that work is unfortunately not made), who derives digital images from general continuous images by means of the 'hit-or-miss'- topology. The approach of Ronse avoids difficulties arising in this more general approach: Ronse's pre-images always have a measure of dimension 2, since they consist of squares. It might be desirable to combine Serra's and Ronse's approaches to derive a more general isomorphism, meaningful for any digitalization of continuous point sets. I believe it is in this more general approach that one should try to settle Ronse's conjecture that to each digital isomorphism there corresponds a continuous image homeomorphism. Ronse develops his ideas gradually, first for binary digital images on a square grid, and then generalizing step by step to a point where his isomorphism becomes applicable to very general discrete point sets with a well-defined adjacency. The isomorphism has four basic defining properties: totality (all 'pixels' of two 'images' can be related), frame preservation (the 'frames' can be related), image preservation (the 'colours' of all 'pixels' of two 'images' can be related) and adjacency preservation (two related 'pixels' have related adjacent neighbours). The image preservation condition makes Ronse's isomorphism as yet unsuited for greytone images, though it can be used for arbitrary N-ary 'images'. Examples applying the isomorphism to polyhedral triangulations and error- correcting codes show the power and generality of the approach. Some remarks: unfortunately, no algorithms are given actually checking whether two given images are isomorphic according to Ronse; at some points the paper lacks in precision; quantifiers are not always used where they should be used and the wording is sometimes ambiguous. But in all, the paper makes clear, easy and stimulating reading.