.NTI1Mw.MjY3NTI5

De Transcription | Bibliothèque patrimoniale numérique Mines ParisTech
Version du 15 mars 2021 à 12:53 par MargauxMines (discuter | contributions)

(diff) ← Version précédente | Voir la version courante (diff) | Version suivante → (diff)
Aller à : navigation, rechercher

ng at arrow See s. Do not fofd this sheet. Image Analysas : a Consumer's guide - F. MEYER, Centre de Gëostatistique et de Morphologie Mathématique, E.N.S.M.P. 35 Rue Saint-Honoré 77305 FONTAINEBLEAU CEDEX Abstract : The last years have seen an explosion of Systems in image analysis. It is hard for the pathologist or the cytologist to make the right choice of equtpment. Ail machines are stupid, and the only valuable thing is the human work put into it. So make your benefit of the work other people have done for you. Chose a method largely used on many Systems and which has proved fertile in many domains and not only for your j spécifie to day's application Mathematical Morphology , to which are to be added the linear convoiutions présent on ail machines is a strong candida¬ te for becoming such a method. The paper illustrâtes a working day of an idéal System : research and diagnostic directed work during the working hours, automatic screening of cervical (or other) ^smears during night. I Introduction : Observing the field of image processing, one has sometimes the feelipg of a large squander of means and money. How many carcases of do-it-yourself image processing devices are in the cupboards of the laboratories ? Is it not a waste of time and effort when a program developed on one machine will never run: on other machines, because both machine and language are so parti cular ? A waste again when a program cannot be transformed to be adapted to another slightly différent problem, because it is a put-up piece of small heuristic tricks specially tailored for a narrowclass of images ..? The common attitude of many scientists (and image processing people act according to rule) is to act in such a way that nobody can spare them. And computers are their best accompli ces. It is so easy to write ni ce large programs, to tailor Systems, certainly doing .an interesting job, but understood only by one person, the designer ? On one si de you will have the image processing devices, software packages and their man-servants. On the other si de, the potential users with growing and diversified needs for image processing. Which advice can we give to the second for making a choice betv/een methods and Systems ? First question : can one afford a machine ? It is certainly easier to find the necessary funding for a machine if it is a multi-purpose machine and if several users share it. The flexibility, power, ease of programmation also ensure that the machine will survive several years without becoming obsolescent. A machine tailored for one précisé application, one tissue, one problem will die soon. An idéal machine will serve for research pur- pose, aid the human diagnosis during work hours, and will do routine work, such as automatic cervical screening during night and week-ends. A good motto for a scientist should be "Never do a work twice, never do a work done by somebody else". Which means that the language used for expres¬ sing the programs should be rich enough and permit the use of the features and algorithms developed by other people or found in the literature. It should also be machine indépendant. Changing a machine should not mean for you 2 years of stérile work of rewriting old programs. My last advice is not to trust anybody's advice. The quality of a tree may be recognized at if s fruits. Therefore, before chosing a method or a device, make your own inquiry on the number and variety of problems this method could solve and more interesting, even if the information is diffi- cult to get, which problems the method could not solve.