Image registration. Principles, tools and methods. (Q660181)

Image registration is the process of finding correspondence between all points in two images of a scene. That is one of the main issues in the application of image processing. This book aims to describe principles, tools and methods in image registration. The author reviews past tools and methods and compares their performances. New tools and methods are introduced and evaluated. The book covers the fundamentals of digital image registration but it does not discuss applications of image registration nor does it discuss characteristics of various types of images. The book is organized into 11 chapters. Chapter 1 is a general introduction to registration of images. It also describes some typical examples of image registration and it ends with a brief history of the subject. Chapter 2 gives an extensive and rigorous definition of several similarity and non-similarity measures. In addition to reviewing classical measures, four new measures are introduced. The discrimination powers of the measures are determined using synthetic and real images, and their sensitivity to noise and image blurring as well as to intensity and geometric differences between images are determined and compared. Chapter 3 is titled ``Point detectors''. Feature points in an image carry critical information about scene structure. In image registration, knowledge of corresponding points in two images is required to spatially align the images. Point detectors appeared in the literature are reviewed. A number of new point detectors is also introduced. The performance of the used detectors under noise and intensity and geometric changes is determined and compared. Finally, a guide to the selection of detectors is provided. Chapter 4 covers topics related to feature extraction. Several image features (122) are reviewed, categorized, and evaluated. Invariance and repeatability properties of that features are tested under image blurring, noise and changes in intensity and geometry. Chapter 5 deals with image descriptors. An image descriptor is a vector containing some information about an image. The components of a descriptor can represent different types of information. Descriptors are needed to determine the correspondence between control points in two images. This chapter describes in detail eight commonly used descriptors. No comparisons are made regarding the efficiency for image registration. The focus in Chapter 6 is to create descriptors from heterogeneous features. Feature selection is the problem of reducing the number of features in a recognition or matching task. As such it arises in regression analysis, independence analysis, discriminant analysis, cluster analysis and classification. In this chapter, several feature selection methods are reviewed and their uses in creation of efficient heterogeneous image descriptors are explored. Chapter 7, titled ``Point pattern matching'', discusses methods that establish correspondence between control points ``outliers'' and control points slightly displaced due to noise and other factors. Point pattern matching algorithms typically use geometric constraints that hold between corresponding points in the images to distinguish correct correspondences from incorrect ones. Chapter 8 is titled ``Robust parameter estimation''. When using image features descriptors to find the correspondence between control points in two images, presence of noise, repeated patterns, geometric and intensity differences between the images can result in some incorrect correspondences. The job of a robust estimator is to identify some or all of the correct correspondences and use their coordinates to determine the transformation parameters. In this chapter, mathematically well-known robust estimators that are not widely used in computer vision and image analysis applications are reviewed. The list of references at the end of the chapter refer to a few recent papers in the area of robust estimation. Chapter 9 is devoted to transformation functions. Such a function uses the coordinates of corresponding control points in two images to estimate the geometric relation between the images, which is then used to transform the geometry of one image to that of the other to spatially align the images. In this chapter, functions most suitable for the registration of images with local geometric differences are examined. The chapter ends by reviewing the properties of various transformation functions. Their performances in registration of images with varying degrees of geometric differences are measured and compared. Chapter 10 is titled ``Image resampling and compositing''. A transformation function maps a point \(a\) in the reference image to a point \(A\) in the sensed image. To estimate the intensity at \(A\), intensities in a neighborhood of \(A\) in the sensed image are used. Different methods to achieve this estimation have been developed. In this chapter, nearest-neighbor, bilinear interpolation, cubic convolution, cubic spline interpolation, and radially symmetric resampling methods are discussed. Chapter 11 is devoted to methods that use the tools of the preceding chapters to register various types of images. Also discussed in this chapter are multiresolution and adaptive registration methods as well as performance evaluation of a registration method. Each chapter concludes with an extensive and somewhat updated list of bibliographic references. This book is a valuable text for students, image analysis software developers, engineers, and researchers who would like to analyze two or more images of a scene.