Image Registration Techniques and Frameworks: A Review

Introduction

Image registration (Lucas & Kanade 1981) is a significant process in several domains including computer vision, medical imaging, biological imaging and brain mapping, military automatic target recognition, and compiling/analyzing satellites’ images and data. It is the act for transforming different sets of data into one coordinate system. This process involves designating one image as a reference image (fixed image), while applying geometric transformations to the other images, thus they align with the reference image. A geometric transformation maps locations in one image to new locations in another image. Consequently, the key step for the image registration process is to determine the correct geometric transformation parameters. Image registration allows the common features comparison in different images (Irani & Peleg 1991; Li et al., 2005). Data may be multiple photographs or data from different sensors; times; depths; or viewpoints. Registration (Christensen & He, 2001) is necessary in order to be able to compare or integrate the data obtained from these different measurements.

There are various algorithms that employed for image registration. The registration techniques can be categorized based on different parameters as follows. The registration methods categories based on image alignment are:

• 1.

  Intensity-based method that compares the images’ intensity patterns using correlation metrics. The registration (Hong & Zhang, 2005) is performed for the entire images or for a part of that image (sub-images), where for sub-images registration the centers of corresponding sub-images are treated as corresponding feature points.

• 2.

  Feature-based method that treats the image features (lines, points and contours) as a parameter to find the correspondence between a numbers of especially distinct points in images (Siu & Lau, 2005; Wahed et al., 2013). Knowing the correspondence between a numbers of points in images, a geometrical transformation is then determined to map the target image to the reference images, thereby establishing point-by-point correspondence between the reference and target images.

The registration methods categories based on the domain are:

• 1.

  Frequency domain method: it determines the transformation parameters for the images registration while working in the transform domain. Such method uses transformations, such as translation, rotation, and scaling. Then, apply phase correlation method to a pair of images, which produces a third image that contains a single peak.

• 2.

  Spatial domain method: it operates in the image domain to match the intensity patterns or features in the images. The corresponding control points (CP) are chosen from the images. When the CP number exceeds the minimum requirements to define the appropriate transformation model, iterative algorithms, such as the RANSAC (Random sample consensus) can be used to robustly estimate the parameters of a particular transformation type (e.g. affine) for the images registration.

The registration methods categories based on the source of images are:

• 1.

  Single methods or mono modal methods: such methods tend to register images (Ashburner & Friston, 2007; Wang et al., 2011) in the same modality acquired by the same scanner/sensor type.

• 2.

  Multi-modality methods: multi-modality registration methods tended to register images acquired by different scanner/sensor types.

The registration methods categories based on the level of automation are: