Search

Browse

• Communities
• Authors
• Year
• Titles
• Subjects

My Space

• Login

Novel and robust methods for the automatic registration of image data

Spanakis Konstantinos

Full record

---

Summary

Image Registration is the process of geometrically transforming two or more images in order for their common points to occupy the same place in space. It is used in many applications such as Medical Imaging, Remote Sensing and Image Stitching. Despite the progress over the last 40 years, there are still unresolved issues: Accuracy, Computational Costs, Convergence to Local Maximums and Automation. The Image Similarity Measure, the Geometric Transformation and the Optimization Method influence these in turn. Mathematical / statistical methods for comparing images have proven to be much more efficient than methods that use image features such as points. In addition, they require minimal (if not all) pre-processing of the images, which renders them automatic. However, due to their using a significant percentage of the images to estimate similarity, they become computationally very expensive, especially when an extensive search for the optimal transformation is required. In the context of this dissertation, extensive research was conducted on Optimization Methods and Image Similarity Measures. Specifically, research was conducted on Elitist Genetic Algorithms as well as new variants of another optimization method known as Harmony Search. In addition, a method was introduced, in order to reduce computational cost, known as the Surrogate Model. Finally, in the context of image similarity estimation, statistical measures based on the statistical deviation of Renyi (Renyi’s Divergence) were compared in order to use the smallest possible percentage of images without reducing the quality of the results.

Available Files

Services