A scheme of colour image retrieval from databases 1 J.M. Fuertes a, * , M. Lucena a ,N.P erez de la Blanca b , J. Chamorro-Martõ Ânez b a Departamento de Inform atica, Escuela Polit ecnica Superior, Universidad de Ja en, 23071 Ja en, Spain b Departamento de Ciencias de la Computaci oneI.A.,E.T.S.deIngenier  õa Inform atica, Universidad de Granada, 18071 Granada, Spain Received 20 October 1999; received in revised form 25 September 2000 Abstract Thispaperpresentsaschemeofimageretrievalfromadatabaseusingqueriespromptedbythecolourandtheshape oftheobjectspresentindierentscenes.Ofthewholeschemeofimageretrieval,wewillfocusattentiononthemodules that allow feature extraction of the component objects from the scenes and the matching of the objects among the dierent images. The de®ned scheme enables the indexing of images by measuring the similarity between the integral objects. Ó 2001 Elsevier Science B.V. All rights reserved. Keywords: Colour image segmentation; Colour spaces; Deformable model; Image retrieving; Content-based retrieval 1. Introduction Over the last decade we have seen an increasing need to save and share multimedia elements of high graphic content. This need has led to the development of mechanisms that allow for an ef- ®cientstorage,accessandretrievalofgraphicsand images organised around large databases. The in- teraction between these databases takes place through queries made up of key images Swain, 1993; Jacobs et al., 1995; Kankanhalli et al., 1996; Gevers and Smeulders, 1999b, 2000; Del Bimbo and Pala, 1997; Del Bimbo et al., 1998; Chang, 1997). Fig. 1 shows the general scheme of image re- trieval. In this paper, we will show the develop- ment of the modules of feature extraction and image matching. In Chang and Fu, 1980), the concept of query by pictorial example was introduced. This shows a retrieval scheme in which the query to the database is made up of one or more images ex- tracted by the system from the characteristics used in the comparison established between the query image and the input data. This is an ad- aptation of the query by example QBE) mecha- nism for alphanumeric data, which laid down the bases of the interfaces of the image retrieval schemes developed later on. In the selection process of the images that ful®l the conditions expressed in the query, it is neces- sary to use some kind of similarity ranking. The segmentation of scenes made up of dierent ob- jects is in the majority of cases ill-de®ned. This is due to the phenomena of occlusion, translation, www.elsevier.nl/locate/patrec Pattern Recognition Letters 22 2001) 323±337 1 Electronic Annexes of Figs. 3, 4, 7±13 available. See http:// www.elsevier.nl/locate/patrec. * Corresponding author. Fax: +34-953002400. E-mail address: jmf@ujaen.es J.M. Fuertes). 0167-8655/01/$ - see front matter Ó 2001 Elsevier Science B.V. All rights reserved. PII:S0167-865500)00128-8