Study of 3D Fundus Pattern Reconstruction and Display from Fundus Images

在本研究中，我们提出一种方法来重建和显示三维Findus图案从一组多个Findus图像在许多方向。眼底是眼球的内底。它是我们身体中唯一能够直接观察血管的地方，并且它对于各种健康状况的诊断非常有用。然而，由于眼底观察是通过眼睛透镜进行的，因此难以定位眼底部分位置的3D位置并测量它们的尺寸。因此，诊断往往强调观察者的主观性。已经提出了一些方法，其中，在相机校准之后，从一对立体图像获得3D眼底显示。但是由于眼底摄像机的视角有限，只能提供一个狭窄的眼底区域的三维重建，本文提出了一种大面积眼底三维重建的方法，并给出了对一些真实的眼底图像的实验结果，发现由于摄像机透镜和人眼透镜等光学系统的影响，眼底图像会产生倾斜。摄像机光学系统可以完全标定，而人眼系统很难标定。在该系统中，眼睛的光学系统被建模为一个单一的透镜，眼底的形状被假定为球形。证明了眼底的真实的像通过单个透镜形成二次曲面。眼底图像的变形反映了二次曲面的参数。因此，从多幅眼底图像中识别出参数，并通过这些参数估计出倾斜度。为了将我们的结果应用于实际应用，还需要一些医生的帮助。但是许多实验结果保证了所开发的方法的实用性。这也是计算机视觉领域的一个新的发展领域。

In this research, we develop a method to reconstruct and display 3D Findus Pattern from a set of multiple Findus Images taken in many directions. The fundus is the inner bottom of an eyeball. It is the only place in our bodies where blood vessels are able to be observed directly, and it is very useful part for a variety of diagnosis on helth conditions. However, it is difficult to locate the 3D positions of the fundus parts position and measure their dimensions, because the fundus observation is made through eye lens. Due to this, the diagnosis is tend to be emphasized on the subjectivity of the viewers.Some methods have been proposed where, after camera calibrations, 3D fundus display was obtained from a pair of stereo images. But it provided only a narrow area of 3D fundus because of the limited view angle of the fundus camera.In this research, we proposed the method of 3D Reconstruction of large area and showed the experimental results for some real fundus images.Findus images are skewed due to optical systems such camera lens and eye lens. The camera optical system can be fully calibrated, while the human eye system is difficult to be calibrated. In the proposed system, the optical system of the eye is modeled with a single lens, and the shape of the fundus is assumed to be sphere. It is proven that the real images of fundus through a single lens form quadratic surfaces. The deformation of fundus images reflects the parameters of the quadratic surfaces. So, from the multiple fundus images, the parameters were identified, and through which the skews were estimated. Then the 3D display in large area was achieved.To put our result to practical use, it still needs some doctors' help. But many experimental results promise the usefulness of the developed method. This is also a development of new area in the field of computer vision.