SIFT Application Research In Medical Image Registration

With the development of medical imaging technology, medical image registration has become a hotspot in the field of medical image processing, it has important significance in clinical diagnosis and treatment of surgical navigation and physician training. SIFT algorithm is a kind of efficient feature detection and matching algorithm, which has good stability for image rotation, scaling, scale and affine transform and light transform, and has been widely applied in medical image registration.Since the part of gray level of the medical image does not have obvious change, and there may be some similar areas, a small quantity of the detected feature points for the SIFT algorithm will cause the matching problem. The SIFT algorithm has some restriction in practical application, such as retinal fundus image as an example, for example, a part of the image is registered effectively, and the other part is unable to realize the registration. Herein, we use retinal fundus images as an example to show the problem. Because of containing a large number of blood vessels in retinal fundus images, the structure information can be extracted, so the bifurcation structure feature description information can robustly realize the registration, but which is not suitable for the non-structural region.In order to carry out the accurate registration for retinal fundus images, a SIFT algorithm based on the bifurcation structure is proposed in the paper. The SIFT description and the bifurcation structure description are adopted in the algorithm. To improve the quantity and the accuracy of the feature points, a control constraint strategy is used in the registration process. A great deal of experiments show that the proposed method has robustness feature description and high registration accuracy in the retinal fundus image registration.