Research On The ICP And The SVD Of The Fundus Images Registration Methods

Medical Image Registration as an interdisciplinary field of research in information science, computer technology and contemporary medical images,has been in clinical diagnosis, treatment, preoperative planning, and so have a wide range of applications. In this paper,we have a comprehensive analysis and summary in medical image registration of the theory, methods and technologies. Due to the physical nature of the fundus, the operator based on rigid registration of medical images. Transformation refers to the so-called rigid have only the axes of rotation and translation.Image transformation is a rigidFundus image is a reflection of changes in hypertensive patients "barometer". Observation of retinal arteries circumstances, Such as arterio-venous diameter, the wall changes, whether or not the phenomenon of cross-oppression and extent of retinal hemorrhage, exudation and the nature of existence,which have some extent to understand the situation of systemic vascular.Fundus fluorescein angiography is commonly used ophthalmic clinical examination techniques, especially for the diagnosis and treatment of fundus plays an important role in choice. However, the application of this technology in the course of a number of various side effects reported.Fundus fluorescein angiography method used contrast agent in patients with fluorescein sodium will have varying degrees of nausea, vomiting, urticaria and even shock, cardiac arrest and other adverse reactions. Therefore, for non-fundus fluorescein angiography study of law is essential. Taking into account the results of fundus measurement accuracy, objectivity, repeatability and high-volume survey of the needs of the fundus, the fundus image analysis, especially with high blood pressure is closely related to changes in the retinal blood vessels, as well as the quantitative analysis of retinal exudative lesions on the proposed extraction the urgent demand.In a comprehensive analysis of fundus images need to be resolved before the rigorous image alignment problem, which is the so-called medical image registration. At present the majority of medical personnel are hit-ray film images, in the light box to carry out observation, the observation of doctors alone will align the pieces of the image space, there are many constraints, and with a strong subjectivity, reliability high, there is unavoidable error, therefore registration of fundus images has become a necessary and very difficult task, then the computer must rely on accurate registration.Image registration can be quickly achieve registration process and has a strong purpose, mainly for retinal fundus images RF (Red Free) images to study some related issues, Include: pre-treatment fundus images, retinal vascular network extraction, the optic nerve head blood vessels feature point extraction and the extract, based on the fundus picture image registration of several major parts: On the fundus image processing for noise reduction, testing a variety of methods are compared. Mainly include the mean filter, low-pass filtering, median filtering, as well as partial differential equations based on the differences to the gradient method. Since the former methods in the main part is to remove high-frequency out, but also to remove the edge of some of the information.Followed by image enhancement of images, we used image enhancement methods are mainly Amendment Act histogram, gray-scale transformation method. Extraction of vascular network using threshold segmentation method, based on the gray - the maximum entropy gradient threshold segmentation method to be the background image of the blood vessels. The average amount of information entropy is a sign, the 20th century, the 80's people began to consider the use of information theory in the concept of entropy threshold selection. Pun first proposed in 1980, maximum a posteriori entropy upper bound method. Johannsen, and in 1982 raised the minimum information Bile interactive method is also used only gray-scale image of the entropy, the method to gray-scale is divided into two parts, and to each other (in the information theory sense) select the minimum threshold.Entropy 1985, Kapur proposed threshold selection method; Abutaleb will be in 1989 and so on Kapur's maximum entropy method of one-dimensional extended to two-dimensional, that is, considering the gray-scale pixel and its neighborhood average gray-scale consisting of two-dimensional histogramPal, such as the 1989 introduction of q-order local image entropy and the concept of conditional entropy, the selected threshold of the largest second-order local entropy law and the maximum entropy method; On this basis the use of optimization algorithm is based on the expansion of Hilditch and corrosion solution, extraction the centerline of the vesselExtraction of the blood vessels cross the center line of the selected cross-points as part of feature points In this paper, extraction of fundus images before optic nerve center of mass of the two cross-points with feature points as a pre-registration, in this based on the ICP algorithm and the SVD matching the combination of methodsComputing has been the use of ICP and the SVD transform iteration operator, iterative meet specific conditions to achieve the ultimate image registration, but also to track the purpose of a particular region.Pre-registration through the first, and then refined through the iterative matching algorithm greatly reduces the precision of the registration number of iterations, the algorithm to improve accuracy and speed, and achieved relatively good results.In the theoretical research and analysis on the basis of the fundus picture in this article achieved the pre-treatment, fundus images to remove noise; Extraction of retinal vascular network through the vascular network of blood vessels and blood vessels cross the center line, the establishment of vascular cross-point topology;Center of mass with optic nerve head and vascular cross-point pre-registration, the last use of the use of ICP algorithm and SVD methods of combining the fine registration.The use of the optic nerve head of the prominent features of the fundus image registration, the use of ICP and computing the SVD to be carried out iterative transform operator, the final iteration to meet specific conditions to achieve precise image registration, but also track the specific purpose of regional Algorithm both in accuracy and speed, there are good results, the realization of the image processing system, consistent with a modular and generic user interface, and has better scalability.