Similarity-based versus template matching-based methodologies for image alignment of polyhedral-like objects under noisy conditions

The image alignment problem has as input a set of different images from a single object or several identical objects. One instance of the object in each image is misaligned relatively to the other in a given image, that is, the objects lie in different positions and orientations. Thus, the problem consists in finding the homogeneous planar transforms for each of those instances. The process is not simple in the case of noisy images. Nevertheless, it is possible to tackle some cases of the image alignment problem using template matching. This work describes two methodologies that were developed to align noisy images of protein quaternary structures obtained by Transmission Electron Microscopy (TEM). These structures can be geometrically described by regular prisms and polyhedra. One of the methods uses the traditional approach for this problem-exploring image similarities. The second one explores the geometric simplicity of the objects being imaged by using template matching in order to solve the problem. The results show that there are cases where the template matching method achieves higher precision and computational efficiency to align that kind of image when compared to traditional ones.