Chernikov,I.
Abstract:
The increasing number of three-dimensional data on the web and huge 3d-model databases in special applications give rise to the problem of its retrieving. At the same time this issue is one of the most popular and complex tasks in computer vision theory. The main challenge is to develop efficient and robust matching algorithm that works with arbitrary polygonal models. One of approaches for model matching is based on global surface descriptors, which represent the shape of the whole 3D model surface in compact and informative manner. We present new global surface descriptor called integral spin image which utilize the concept of popular local surface descriptors spin images. We also propose special 3D model normalization method for integral spin image estimation. We show that the new model matching algorithm based on integral spin images provides favorable shape similarity measure and can be successfully used in 3D model search systems.