Content Based Thermal Images Retrieval Hern´ an Dar´ ıo Ben´ ıtez and Gloria In´ es Alvarez Pontificia Universidad Javeriana Cali, Calle 18 No. 118-250, Cali, Colombia {hbenitez,galvarez}@javerianacali.edu.co http://www.javerianacali.edu.co Abstract. INDT (Infrared Nondestructive Testing) inspections produce large sets of thermal images. The thermal images obtained must be stored to do comparisons with the results of previous and future inspections gen- erating considerable amounts of data to be analyzed. In this paper, we present a CBIR system based on thermal features and spatial relations among defects in the thermal image to measure the similarity between query and database images affected by non uniform heating and taken from anisotropic material samples. Keywords: Content based image retrieval, spatial relationships, infrared nondestructive testing, thermal contrast. 1 Introduction Infrared Nonsdestructive Testing (INDT) is a nondestructive evaluation tech- nique in which the specimen surface is thermally stimulated to produce a tem- perature difference between sound (free of defects) areas and eventual defective regions. This technique is fast, non-invasive and is used in several industry fields such as automotive and aircraft industry [1]. In aircfrat industry, INDT is well suited for in-service damage detection in CFRP (Carbon Fiber Reinforced Plas- tic) [2] structures since it is well suited to detect shallow delaminations and can be deployed directly on site. INDT inspections produce large repositories of digital thermal images. The thermal images obtained must be stored to do comparisons with the results of previous and future inspections, generating considerable amounts of data to be analyzed. Effective extraction of visual and temperature features is needed to provide meaningful access to visual data and retrieve thermal images based not on textual annotations but on visual features [3]. We propose the application of content-based image retrieval (CBIR) for thermal images obtained from INDT inspections. CBIR allows the retrieval of relevant images based on a pre-defined similarity measure between image features. For example, this type of system could be used to measure the true temperature on Printed Circuit Boards (PCB) that are produced in a large scale by the electronic industry. In this paper we present the thermal image representation based on thermal constrast and spa- tial relations between defects found in the image. Then, we describe the image A. Elmoataz et al. (Eds.): ICISP 2010, LNCS 6134, pp. 479–487, 2010. c  Springer-Verlag Berlin Heidelberg 2010