Image analysis of multiphased ceramics P. Belhomme, D. Houivet, W. Lecluse, J.M. Haussonne* Laboratoire Universitaire des Sciences Applique ´es de Cherbourg, Universite ´ de Caen Basse Normandie, LUSAC, Site Universitaire de Cherbourg, BP 78, 50130 Octeville, France Received 4 September 2000; received in revised form 20 November 2000; accepted 5 December 2000 Abstract Inthisstudy,wefocusontherecognitionandcharacterizationofthedifferentphasespresentina(Zr,Sn)TiO 4 ceramic(withsome amountsofLa 2 O 3 andNiOassinteringaids).Ourworkdealswiththestudyofsegmentationtechniquesthatautomaticallyextracteach ofthephases(ZST,TiO 2 ,La 2/3 TiO 3 ,...)fromascanningelectronicmicroscopeimageofapolishedsectionoftheceramic.Themain difficultyencounteredatthisstageoftheworkisthattheirpropertiesintermsofgraylevelsareveryclosetogether,thus,imposingthe combinationofsimpleimageprocessingoperators(suchasthresholding)withadvancedmethodsbasedon region growing processes. # 2001 Elsevier Science Ltd. All rights reserved. Keywords: Electron microscopy; Image processing; Microstructure-final 1. Introduction The microwave characteristics of (Zr,Sn)TiO 4 dielec- tricceramicsforfiltersorresonatorsarehighlydependent on the process. More precisely, these ceramics are multi- phased, the nature, amount and shape of the different phases depending on the whole steps of the ceramic pro- cess. 1 So, it is important to be able to evaluate and quantifyinanautomaticwaytheinfluenceofthediffer- entprocessparametersonthesizeoftheceramicgrains, ontheamountandrepartitionofthedifferentsecondary phases, etc. It is typically in this kind of situation that image processing appears to be an interesting tool, thanks to the great performances available with PCs. 2. Materials SEM micrographs of ceramics from the composition diagramZrO 2 –SnO 2 –TiO 2 (withsomeamountsofLa 2 O 3 and NiO as sintering aids) were acquired at 500 magnifi- cation with a secondary electrons detector (acceleration voltage: 20 kV, working distance: 25 mm). Ceramics have previously been polished and sputtered with gold. The image size is 512368, each pixel being coded by one byte (256 gray levels possible). Fig. 1 shows a typi- cal image to be processed, which was obtained after a smoothing step by a gaussian filter (size 3,  ¼ 5) for reducing acquisition noise. Considering the gray level distribution (see histogram Fig. 2), the operator has to center the ZST peak on the value 128, then to tune acquisition devices in order to spread out the image gray levels on the whole available range. A slight saturation is thus encountered for hole side reflections. In the presented example five classes have to be extracted, which are respectively:  in black: holes;  in dark gray: TiO 2 phase;  in gray: ZST phase (corresponding to the image background);  in light gray: La 2/3 TiO 3 phase;  in white: hole side reflections. In this work, we have used two image processing softwares. The first one is a commercial toolbox based on a graphical user interface (GUI) for rapid testing of the operators to be used. The other one, developed in a laboratory of our University, allows the creation of an executable file leading to the complete automation of 0955-2219/01/$ - see front matter # 2001 Elsevier Science Ltd. All rights reserved. PII: S0955-2219(01)00191-1 Journal of the European Ceramic Society 21 (2001) 2149–2151 www.elsevier.com/locate/jeurceramsoc * Correspondingauthor.Tel.:+33-02-3301-4214;fax:+33-02-3301- 4201. E-mail address: jmhaussonne.lusac@chbg.unicaen.fr (J.M. Haus- sonne).