112 Rakesh Prabhu, A. M. Khan International Journal of Engineering Technology Science and Research IJETSR www.ijetsr.com ISSN 2394 – 3386 Volume 4, Issue 9 September 2017 A Study of Bio-medical Image Registration Algorithms Rakesh Prabhu Research Scholar Department of Electronics Mangalore University , Mangalore, India. A. M. Khan Professor Department of Electronics Mangalore University , Mangalore, India Abstract— Bio-medical Image Registration has greater impact on image analysis. Image Registration technique helps to align two images, which uses intensity of pixels as prime factor for registration. The images used for registration can be either mono-modal or multi-modal. The degradation of image is frequently observed during image acquisition itself, which affects registration considerably. The performance of algorithms should be independent of image degradations. The most commonly used techniques for similarity measures to perform image registration include Sum of Absolute Difference (SAD), Sum of Squared Differences (SSD), and Normalized Cross Correlation (NCC) methods. In this paper a comparative study is done to obtain better algorithm. A set of images were taken for the study and were modified by varying contrast and relative shift. The obtained results show that SAD method is best suited to perform registration of mono-modal images. Keywords— Image Registration, SAD, SSD, NCC I. INTRODUCTION Image Registration plays major role in all applications of medical image analysis and is a basic need to perform image fusion. Registration technique helps for long period abnormality monitoring to determine changes over period [1,2]. Image registration is an optimization procedure that uses similarity measure to find the optimal alignment of two images. Image registration if done on images acquired using same modality then it is called mono- modal registration else it is called multi-modal registration. The accuracy of image registration process depends on geometric transformation and optimization algorithm based on similarity measure [3]. Selection of similarity measure depends on the type of registration used and on intra-/inter-subject perspective. As a resultmany image similarity measuring techniques have been developed over the time. The most commonly used similarity measures include Sum of Absolute Difference (SAD), Sum of Squared Differences (SSD), and Normalized Cross Correlation (CC) methods [4, 13]. In the study of registration algorithm, Computed Tomography (CT) image is used and wide framework is considered to perform evaluation of Image Registration techniques. The rest of the paper is organized as follows. Section II, explains suitable theory of the registration algorithms, Section III includes experiment and results of different similarity-measure based on image registration techniques, and section IV brings conclusion of the study. II. IMAGE REGISTRATION Image registration is a process of over lapping two or more images of scene taken at different time or from different view point and/or obtained by different sensor [6, 10]. In this section, image registration approach is discussed in detail. First, we introduce geometrical transformation and then discuss different similarity measures utilized in our study. Further, a Levenberg-Marquardt (LM) optimization algorithm is also presented. Finally, the registration algorithm steps are summarized for proposed study. A. The geometrical transformation model Image registration uses geometrical transformation to align a references imageI1(x) and source image I2(x). Geometrical transformation modifies the spatial relationship between pixels in an image. The geometrical transformation is classified in to rigid, affine, and non-rigid based on their degree of freedom. Affine geometrical transform is one which has higher degree of freedom with respect to rigid transformation and lower degree of freedom with respect to non-rigid transformation [5, 6]. Affine