Image registration using robust M-estimators K.V. Arya a, * , P. Gupta b , P.K. Kalra c , P. Mitra d a Department of Computer Science & Information Technology, M. J. P. Rohilkhand University, Bareilly 243006, India b Department of Computer Science & Engineering, Indian Institute of Technology Kanpur, Kanpur 208016, India c Department of Electrical Engineering, Indian Institute of Technology Kanpur, Kanpur 208016, India d Department of Computer Science & Engineering, Indian Institute of Technology Kharagpur, Kharagpur 721302, India Received 1 July 2005; received in revised form 28 April 2007 Available online 24 May 2007 Communicated by M. Kamel Abstract In this paper, a method for robust image registration based on M-estimator Correlation Coefficient (MCC) is presented. A real valued correlation mask function is computed using Huber and Tukey’s robust statistics and is used as a similarity measure for registering image windows. The mask function suppresses the influence of outlier points and makes the registration algorithm robust to noisy pixels, brightness fluctuations and presence of occluding objects. The superiority of the proposed algorithm, in terms of registration perfor- mance and computation time is demonstrated through experimental studies on different types of real world images. Ó 2007 Elsevier B.V. All rights reserved. Keywords: Robust statistics; M-estimator; Huber statistics; Tukey statistics; Normalized cross-correlation; Influence function 1. Introduction Many image processing applications often need to com- pare or combine information given by multiple images. To perform this task, image registration is one of the funda- mental steps. Image registration is the process of determin- ing correspondence between all the points in two images of the same scene. One of the images used for registration is kept unchanged and is referred as reference or template image while the other one can be warped, and is called the sensed or target image. Template matching is a popular method for registering objects, symbols, characters and faces due to the simplicity of implementation. Template matching is the process of finding location of a sub-image, called template image, inside a given image. It involves determining the similarities between a given template and windows of the same size in the target image, and then identifying the window that pro- duces the highest similarity measure. Registration in real world images involves problems of noisy environment and shadow or occlusion in the image scene. Robustness is an important property required for successful registra- tion in above environments. Many feature-based robust registration methods have been proposed in literature (Brown, 1992; Dai and Khor- ram, 1999; Ghaffary and Sawechuk, 1983; Zitova and Flus- ser, 2003). These methods try to match image features, e.g., lines, corners, contours, between the target and the refer- ence image. Clustering technique presented by Goshtasby et al. (1986) and Stockman et al. (1982) attempt to match points connected by line segments. Barrow et al. (1977) introduces the chamfer matching for image registration, where line features detected in the two image are matched by minimizing the distance between them. Borgefors (1988) 0167-8655/$ - see front matter Ó 2007 Elsevier B.V. All rights reserved. doi:10.1016/j.patrec.2007.05.006 * Corresponding author. Tel.: +91 581 2520003; fax: +91 581 2528384. E-mail addresses: kvarya@gmail.com (K.V. Arya), pg@iitk.ac.in (P. Gupta), kalra@iitk.ac.in (P.K. Kalra), pabitra@cse.iitkgp.ernet.in (P. Mitra). www.elsevier.com/locate/patrec Pattern Recognition Letters 28 (2007) 1957–1968