Multiscale gradient based digital radiography image registration
Dorde S. Starcevic, Vladimir S. Ostojic, Vladimir S. Petrovic · 2017
Two or more digital radiography images are stitched together after acquisition to represent the long bone anatomies. To perform the stitching, optimal translation between these composing images must be determined. We propose a method that combines anatomical and reference structure alignment. The proposed method is multiscale for faster computation, and gradient based to overcome intensity differences between images. Evaluation was performed on a dataset of medical and non-medical digital radiography image pairs. The average absolute difference of manually and automatically determined translation parameters is 1.29 pixels in the direction of the stitching and 2.29 pixels in the direction perpendicular to the direction of the stitching.