Automated detection of particles, clusters and islands in scanning probe microscopy images M.J.J. Jak a,1 , C. Konstapel a , A. van Kreuningen a , J. Verhoeven a, * , R. van Gastel b , J.W.M. Frenken b a FOM-Institute for Atomic and Molecular Physics, Kruislaan 407, 1098 SJ Amsterdam, Netherlands b Kamerlingh Onnes Laboratory, Leiden University, P.O. Box 9504, 2300 RA Leiden, Netherlands Received 24 January 2001; accepted for publication 4 August 2001 Abstract In order to obtain quantitative information from scanning tunnelling microscopy STM) images, image processing andpatternrecognitiontechniquesareveryvaluabletools.Wedevelopedanalgorithmwhichautomaticallydetermines thepositionsandsizesofsmallparticlesandothernanostructuresinSTMandatomicforcemicroscopyAFM)images. This algorithm has been tested and used both in the study of Pd nanoparticles supported on TiO 2 andinthestudyof diusingInatomsembeddedinaCusurface.FirsttheoriginalSTMimageis®lteredinordertoobtainanimageofthe background. Subtracting this `background' image from the original image eliminates the height variations in the substrate, such as atomic steps. The particles can then be found by discrimination with respect to a threshold height. Once the particles are located, their exact position and size are determined and used for further analysis. Ó 2001 Elsevier Science B.V. All rights reserved. Keywords: Scanning tunneling microscopy; Clusters 1. Introduction Scanning probe microscopy SPM) techniques such as scanning tunnelling microscopy STM) andatomicforcemicroscopyAFM)havebecome important tools in the study of adatoms, islands, clusters, and small supported particles. Images obtained from these studies can contain a wealth of both qualitative and quantitative information. A detailed quantitative analysis of the images is often required to fully reveal the underlying physics. The extraction of reliable, numerical data from an image can, however, be a tedious task. For example, the average density of a particular type of particle on the surface can simply be de- termined by counting the number of those parti- cles in a series of images. Usually, counting is performed `by hand' and the statistics is limited. Therefore it is very useful to develop automatic procedures to analyse the STM or AFM images and extract the relevant numbers. In this way much larger data sets can be handled, more com- plex analyses can be performed, and greater ac- curacy can be achieved. Surface Science 494 2001) 43±52 www.elsevier.com/locate/susc * Corresponding author. Tel.: +31-20-6081234; fax: +31-20- 6684106. E-mail address: verhoeven@amolf.nl J. Verhoeven). 1 Present address: Philips Research Laboratories, Prof. Hol- stlaan 4, 5656 AA Eindhoven, Netherlands. 0039-6028/01/$ - see front matter Ó 2001 Elsevier Science B.V. All rights reserved. PII:S0039-602801)01487-X