Geometrically constrained sub-pixel disparity estimation from stereo images of the retinal fundus

Abstract

The aim of this study is to propose a precise disparity estimation process between retinal fundus stereoscopic images to assist during eye diagnosis and view generation for visualization. We propose a computer-vision-based solution that allows, from stereo images, the extraction of precise disparity information that could be used for clinical parameters estimation and/or the generation of multi-viewpoint images of retinal fundus. The proposed solution is based on a sub-pixel disparity estimation algorithm that takes into consideration the spherical shape of the retinal fundus. A sub-pixel approach is adopted in order to achieve high precision disparity estimation. Moreover, the a priori knowledge of the fundus geometric shape provides useful information for the regularization of the high precision disparity estimation process. Stereo images, with known ground truth, are used to evaluate the proposed algorithm and to demonstrate the high precision of estimated disparity in our methods.

Real‐time fundus reconstruction and intraocular mapping using an ophthalmic endoscope

“…Regarding the eyeball estimation, we estimate the object eyeball as a sphere, as in some previous studies. [10][11][12]20 The formation of the eye ball function is represented by Equation (11).…”

“…developed a constrained optimisation procedure to estimate the retina curvature considering the geometric shape of the human retina and lens distortions. Kharboutly et al 12 . proposed a precise disparity estimation process between retinal fundus stereoscopic images to assist during eye diagnosis and view generation for visualisation.…”

“…Chanwimaluang et al 11 developed a constrained optimisation procedure to estimate the retina curvature considering the geometric shape of the human retina and lens distortions. Kharboutly et al 12 proposed a precise disparity estimation process between retinal fundus stereoscopic images to assist during eye diagnosis and view generation for visualisation. For local matching, Laliberte et al 13 reconstructed the 3D surface information of a local fundus part using a sequence of fluorescein angiograms.…”