In clinical practice, digital subtraction angiography (DSA) is a powerful technique for the visualization of blood vessels in a sequence of X-ray images. With this technique, a sequence of 2D digital X-ray projection images is acquired to show the passage of injected contrast material through the vessels of interest. In the images that show opacified vessels (often referred to as contrast images), background structures are largely removed by subtracting an image acquired prior to injection (called the mask image). The major problem encountered in DSA images is the presence of motion artifacts which arise from the misalignment of mask image and contrast images in the sequence and reduce the diagnostic value of DSA images. To cope with this problem, registration of the mask and contrast images is required prior to subtracting the images. Image registration is the process of spatially aligning two images of the same scene taken at different times, from different viewpoints, and/or by different sensors. During the past thirty years, enormous efforts have been spent on this subject and great progresses have been achieved. However, because of the factors such as complicated non-rigid motion of the tissues inside human body and local dissimilarities caused by contrasted vessels, registration of angiographic images especially coronary angiographic images (concerned angiographic images in this thesis) is very difficult. Most of image registration techniques that have been developed for application in DSA are point-based techniques. In these techniques, the registration process is carried out in three steps: selection of a set control point from mask or contrast image, calculation of displacement of selected control points, and warping of mask image. In this thesis, these steps are studied in details and after that, performance of three registration algorithms evaluated experimentally. Then, two new approaches to the registration of digital angiography images are proposed. First proposed approach is a point-based registration method in which control points are selected using Harris corner detector and edge-based method and entropy is exploited as similarity measure in template matching that for reducing the computational cost of template matching procedure, a hill-climbing approach is used for optimization. The final correction is performed by warping of mask image using the multilevel B-spline interpolation function. In the second proposed registration method, a multiresolution search strategy and least-squares optimization are used. The experimental results show that the proposed methods outperform the previously presented DSA image registration approaches, particularly, first method that is Key Words: Digital subtraction angiography, image registration, motion artifacts, template matching, similarity measure.
