Based Multi-scale Features Of Phase X-ray Image Stitching Method

To solve the problem that the size of the part to be examined is much larger than the area of detector of imaging (such as the spine and the entire lower limb, etc.), the part often can be divided into several sub-parts to image, and then mosaic these images from those sub-parts together according to certain rules, to satisfy the clinical diagnosis and the surgery planning. Image mosaic technique plays an important role in scoliosis and lower limb deformities and other diseases.Phase correlation is sensitive to light intensity, so an image stitching algorithm based on multi-scale phase correlation is presented in this paper, which solves the splicing problems of the X-ray images when noise and intensity inhomogeneity exist. Firstly, the image to be stitched is transformed to rough scale space, and processed using contour extraction. Secondly, translational parameters for initial splicing are acquired with phase correlation, which will be used to determine the potential overlapped regions. Then the translational parameters are updated with phase correlation in small scale space. Thirdly, the affection of the splice is evaluated using the local maxima of multiple pulse functions.Finally, the stitching algorithm based on multi-resolution is used to complete the process. When used to splice static mosaic panorama image, the algorithm is proved to be effective. Compared with classical phase correlation, the algorithm introduced in this paper is independent of scene, insensitive to noise, robust and also effective when there are severe intensity inhomogeneity and geometric distortion.