3D Reconstruction Technique Of Retina Based On Monocular Vision

Image processing has been widely used in medical field, doctors can according to the medical image to the patient's condition for non trauma analysis. The retina is the eye very important part, some diseases such as diabetes, hypertension, macular disease, will have a effect on the retina. According to the subtle changes in the retina, the diagnosis of these diseases, so the retinal image for the treatment of a variety of ocular fundus disease provides a reliable basis for diagnosis. 3D reconstruction of retinal, can better achieve the results of retinal visualization, so that the doctor for some fundus disease diagnosis is not just rely on the past experience, the diagnosis is more accurate and more reliable.This paper is based on the 3D reconstruction of retinal: retinal image acquisition, camera calibration, retinal image preprocessing, feature extraction and matching of retinal image, 3D reconstruction and Realization of 3D reconstruction system. The main research contents and innovations are as follows:(1) Analysis of analyzes the current background and practical significance of the retina, studies and analyzes the research status at home and abroad, expounds the background and significance of the topic, makes a detailed arrangement of the chapter and the research content of the thesis.(2) According to the characteristics of obtaining retinal images, the structure of human eye and retinal structure were studied, and the imaging model of the camera was analyzed.(3) According to the characteristics of fundus camera calibration, the current calibration methods are studied and analyzed. By comparing the calibration method, calibrated by the method of Jean-Yves Bouguet, high precision and small error calibration results; Image preprocessing is proposed by using contrast limited adaptive histogram equalization(CLAHE), after processing the image contrast enhanced.(4) For retinal feature extraction and matching characteristics, analysis the current image matching processing method, comparing proposed retinal image feature extracted by SIFT feature extraction algorithm, matching feature points using iterative nearest neighbor matching method matching, the matching results purification using RANSAC algorithm purification, higher accuracy.(5) For the 3D reconstruction process of the retina based on monocular vision, the basic matrix is obtained by the method of plane and parallax, get the space point data; 4 pass bundle adjustment method is used to get the precise camera position and sparse point cloud, PMVS algorithm method is used to obtain dense point cloud data, and finally to poisson reconstruction and texture mapping for 3D reconstruction of the retina.(6) For the 3D reconstruction process of the retina based on monocular vision, use MATLAB design of 3D reconstruction system of retina.