Geometry Image Representation Of Point Cloud Surfaces And Its Application

With the rapid development of digitizing devices & their application in CAD and Computer Graphics, point cloud surface is receiving more and more attention and gradually becoming one of the main 3D model representations. It is now widely used in reverse engineering, animation, and mechanic manufacture fields. Among the model representation methods, point cloud surface is a direct and initial one. However, a detailed point cloud surface usually causes large data size and takes a long time to transmit through the network.This thesis presents a novel approach for representing 3D point cloud surfaces using geometry images. 3D points are converted to images based on spherical polar coordinates, and point cloud surface information such as location, normal and color are stored in two-dimensional images. To create geometry images of point cloud surfaces, we first determine the center of the polar coordinates and convert the Cartesian coordinates [x,y,z] to polar ones r, Coordinate is encoded as the location and r is encoded as the gray value in the image. To represent point cloud surfaces with high complexity, we propose using multi-layered images. From the point cloud surfaces, multi-layered geometry images are created which can represent the original models accurately very much.The compounded geometry image representation of point cloud surfaces, is also based on the former transform method, where the difference lies: Take one of coordinates as the gray value and another two as location respectively, three different geometry images can be created. Three gray images are combined into a compounded color image.This thesis proposes a progressive transmission of point cloud surfaces on the network based on geometry image representation. Based on the characteristics of the geometry images obtained, an enhanced pyramid image structure is developed toimprove the performance of progressive image transmission. From the progressively transmitted image codes, the client side can rebuild multi-resolution images from which point cloud surface series with different level of details can be rebuilt: the initial point cloud surface rebuilt from transmitted images has the gross structure of the original 3D model, and step by step, the rebuilt point cloud surfaces have more and more details until all the geometry images are transmitted out. In this manner, the original 3D model is rebuilt completely on the client side, and this achieves high performance progressive network transmission of point cloud surfaces.We also present compression and simplification methods of point cloud surfaces based on the geometry image representation.This thesis is organized as follows. Chapter 1 introduces related concept and previous work. Chapter 2 introduces Cartesian-Polar transform model and multi-layered geometry image representation method. Chapter 3 introduces compounded geometry image representation method. Chapter 4 introduces progressive transmission of point cloud surfaces based on geometry image representation. Chapter 5 introduces simplification and compression of point cloud surfaces. And conclusions are drawn in Chapter 6.