Multimodal Image Registration Using Hybrid Transformations

Feature selection is the fundamental step in image registration. Various tasks such as feature extraction, detection are based on feature based approach. In the current paper we are going to discuss about our technique that is hybrid of Local affine and thin plate spline. An automatic edge detection method to achieve the correct edge map is put forward to dealing with image registration with affine transformation for the better image registration. Registration algorithms compute transformations to set correspondence between the two images. The purpose of this paper is to provide a comprehensive comparison of the existing literature available on Image registration methods with proposed technique. Key words— Image registration, Feature matching, Local Affine transformations, Correlation, RANSAC, Thin-plate spline method. I.INTRODUCTION Image registration is the initial step in almost all image processing applications. Image registration is the process of matching two or more images of the same target by different time, or from different sensors or by different perspectives. It is the establishment of various geometrical transformations that will align points in one view of the source image with corresponding points in another view of that image or the reference image [2]. Image registration can be defined as the determination of a one-to-one mapping between the coordinates in one space and the ones in other, such that points in the two spaces that correspond to the same anatomical point are mapped to each other.  Image Registration is a process of finding an optimal transformation between two images.  Sometimes also known as “Spatial Normalization” (SPM). The present difference between the images is due to different imaging conditions. The main objective of image registration is to combine data from different sources accurately and without any redundancy, because the data contained by sources depends upon the method of acquisition. Basically we can define method of acquisition in following ways: A. Multiview analysis (Different viewpoints): Images of the same scene may be acquired from the different viewpoints. In this, images may differ in translation, rotation, and scaling, more complex transformations mainly due to camera positions [5]. Examples include: computer vision shape recovery. B. Multitemporal analysis (Different times): Images of the same scene may be acquired at different times or under different lightning conditions may be [5]. Our aim is to identify and evaluate changes in the scene which appeared between the consecutive images acquisitions. Examples includes: Medical image monitoring, remote sensing. C. Multimodal analysis (Different sensors): Images are acquired by different types of sensors [5]. Here, the aim is to integrate the information from two different sources and then to obtain more representation detail. D. Scene to model registration: Images of a scene and model of a scene are registered [5]. The aim is to localize the acquired image in the model and to compare them. Examples include: Medical imaging.  Image registration can be of rigid or of non-rigid registration. In rigid registration, only rotation and translation are applied for the spatial transformation. It is usually necessary to align the images from the same subject free of any deformation between the acquisitions of the image or as a pre-alignment for methods of the registration with higher degrees of freedom [11].  Non-rigid image registration is a natural extension to rigid registration, allowing also deformations in an image, in order to achieve a good match. This is necessary in all instances of a patient’s motion within an image, such as either in respiratory or in heart motion. It is also necessary if the datasets from different patients must be registered, like in atlas registration applications [14]. Fig 1. Registered source, obtained by matching source and target image by correspondence map Most of the Registration methods consist of the following four steps: Fig2. Block diagram of image registration. International Journal of Technical Research and Applications e-ISSN: 2320-8163, www.ijtra.com Volume 2, Issue 4 (July-Aug 2014), PP. 39-43 40 | P a g e  Feature detection: Edges, contours, line Intersection; corners etc. are manually or automatically detected.  Feature matching: In this step, the common features that are detected in the sensed image and those detected in the reference image are matched.  Transform model estimation: The type and parameters of the so-called mapping function should be chosen according to the prior knowledge, aligning the sensed image with the respective reference image, are then estimated.  Image re-sampling and transformation: The registered image needs to be resampling, after determining the transformation parameters. Image registration can also be classified on the basis of the dimensionality, nature of the registration basis, nature of transformation, domain of transformation [4], interaction, and optimization procedure, modalities involved in the registration, subject and object. We can also explain Image registration with the help of figure3: Fig 3. Image registration Here, firstly we have two sets of medical images that we need to register. Then, preprocessing of both sates take place which includes, determining the image size, pixel spacing, noise present, etc. Then, registration process is started with the help of techniques based on mutual information or interpolation etc. then registered image is reconstructed and we have the final registered image. II.LOCAL AFFINE Local affine transformation is used for withinsubject registration when there is global gross over distortion [1]. The affine transformation preserves the straightness of lines, and hence, the planarity of the surfaces and it also preserves parallelism between the lines, but it angles between lines may change.  All rigid transformations can be treated as affine transformations, but all affine transformations are not rigid transformations.  An affine transformation involves rotation, shearing, translation, scaling as shown in figure 4. Fig 4. Affine Transformations [9]  Affine transformations are represented in homogeneous coordinates because the transformation of point A by any affine transformations can be expressed by the multiplication of a 3*3 matrix and a 3*1 point vector [9]. III.PROPOSED WORK: We have proposed a hybrid method of Local Affine and Thin Plate Spline for effective Image Registration with comparatively less computation time and with more accuracy. The methodology of work will start with the overview of image registration algorithms. The results of various algorithms will then be interpreted on the ground of various quality metrics. Thus, our methodology for implementing the objectives can be summarized as follows:1. To focus on the feature detection so that we decide which algorithm to be applied. 2. Based on the Detection we will apply the hybrid Local Affine transformation and Thin Plate Spline (TPS) for Image Registration. 3. This will increase the efficiency of the registration method and quality in the Image. Fig.5. Overview of Hybrid Technology Consists of Local Affine transformation and Thin Plate Spline (TPS) International Journal of Technical Research and Applications e-ISSN: 2320-8163, www.ijtra.com Volume 2, Issue 4 (July-Aug 2014), PP. 39-43 41 | P a g e IV.QUALITY METRICS Various quality measures are there to compare different registration algorithms. Some of these are discussed below: