The main part of presented algorithm is sequential homotopic thinning transformation, “invented” by a group of mathematicians, including F. Meyer (1976), Ch. Lantuejoul (1978), and headed by J.Serra, from Center for Mathematical Morphology, Fontainebleau, France. Many transformations based on thinning were described and published, including “SKIZ”, watersheds by flooding, and markers-controlled watersheds.
I personally learned about MM from “Summer School in Mathematical Morphology” (Fontainebleau, June 1981), and was very fortunate to be able to show some of these results to Professor Serra in 1983, when he was in Moscow and visited our Lab. He was very happy to see how his ideas were implemented for the study of human brain. As far as I know, it was the first implementation of fully automated image processing technology for microscopy images of human cortex.
Practical results of the application of the algorithm presented below were published for the first time in 1984 (V. Istomin, M. Shkliarov, “Automatic analysis of cerebral cortex using television image analyzer”, Korsakov’s Bull. of Neuropathology and Psychiatry, v.7, pp.969-974), and in 1987 in USSR (Istomin, VV. Automatic morphocorticography of human brain (Technology development and application results). Dissertation of the Candidate of Medical Sciences. (Автоматическая морфокортикография мозга человека (Разработка и результаты применения метода). Диссертация на соискание учёной степени кандидата медининских наук.), P.Lumumba’s University of Friendship of the Nations, Moscow, 1987.). Later it was published in 1990 in Europe (V. Istomin, K. Fiedler. Die Automatische Morphocorticographie – eine method zum Erfasses von Veranderungen der Zytoarchitektonik des menschlichen Cortex Cerebri. Bild und Ton, Jan, 1990, 11-14).
As one can imagine, compared to present time, “the state of the art” of microscopy image processing was quite different thirty-plus years ago One of many problems we had to face was extremely slow performance of digital skeleton algorithm. So, it had to be replaced for practical application by segmentation based on mean gradient distribution. Obviously, presented re-implementation of the segmentation is developed using quite different language and more powerful library (MAMBA-Image, CMM), not to mention much faster and more powerful computer :-).