Synthesis of stereoscopic views from monocular endoscopic videos

Recent studies have shown that 3D imaging provides some unique advantages over traditional 2D imaging for minimal invasive surgery. However, most existing endoscopes still use single-lens cameras, and the use of dual-lens 3D imaging techniques is still limited. This paper proposes an approach to enabling 3D imaging from a single-lens endoscope by automatically synthesizing stereoscopic views from monocular images captured by the endoscope. We first formulate the problem by introducing the notion of normalized disparity, based on which we show that affine reconstruction is sufficient for stereoscopic view synthesis. With this formulation and exploiting other domain-specific constraints, we then propose a robust structure-from-motion algorithm for a sparse set of feature points and a fast, linear interpretation algorithm for creating a dense disparity field for synthesizing stereoscopic views from original monocular video. Both synthetic images and real endoscopic videos are used to evaluate the proposed method. The results demonstrate the feasibility and effectiveness of the proposed method.