Plane-Fitting Robust Registration for Complex 3D Models

In surgery planning, forensic and archeology, there is a need to perform analysis and synthesis of complex 3D models. One common first step of 3D model analysis and synthesis is to register a reference model to a target model using similarity transformation. In practice, the models usually contain noise and outliers, and are sometimes incomplete. These facts make the 3D similarity registration challenging. Existing similarity registration methods such as Iterative Closest Point algorithm (ICP) [1] and Fractional Iterative Closest Point algorithm (FICP) [2] are misled by the outliers and are not able to register these models properly. This paper presents a plane-fitting registration algorithm that is more robust than existing registration algorithms. It achieves its robustness by ensuring that the symmetric plane of the reference model is registered to the planar landmarks of the target model. Experiments on patients’ skull models show that the proposed algorithm is robust, accurate and efficient in registering complex models.