Hole Filling in 3D Scanned Models base on 2D Image Inpainting

• Main Authors: Shu-Chin Lin, 林書勤
• Other Authors: Timothy K. Shih
• Format: Others
• Language: en_US
• Published: 2017
• Online Access: http://ndltd.ncl.edu.tw/handle/50607047512932155822

碩士 === 國立中央大學 === 資訊工程學系 === 105 === Nowadays 3D model scanning is widely used for many purposes. This kind of technique is used extensively in many applications such as entertainment industry in the production of movies and video games, industrial design or prototyping. Before scanning technique was invented, even common object like vase needed to be used in virtual environment; digital artists had to draw the entire model from scratch. With scanning technique, artists now can use scanned model as blueprint and beautify the details of model, then the model is available to be used. Although scanning technique has become much advanced then earliest, users may still obtain damaged model due to unfamiliar with scan technique. Unfortunately, the reasons of causing incomplete scanned result are neither obvious nor simple. There are so many factors in real environment such as surface properties, occlusions or light pollution that can cause the result model to be incomplete or damaged. Then we need to repair the model by doing hole-filling and surface reconstruction. In this paper, we introduce an approach to fill holes in 3D mesh by using 2D Image inpainting technique. Image inpainting is the process of reconstructing lost or deteriorated parts of image. Traditionally, inpainting has been done by professional restorers. However, image inpainting has become an automatic process that is performed on digital images. There are also different inpainting methods for different purpose. Structural inpainting uses geometric approaches for filling in the missing information in the region, which should be inpainted. Texture inpainting fills inpainted region with patches from known region in order to recreate texture. By rendering model on screen, we can obtain depth and color image of the model. The value of depth image is actually the z distance between vertices and near plane in orthographic projection, we render z distance in vector form to create depth image. We create image of depth value to represent depth information and image of color to represent color of surfaces, then inpaint those images and use information of result images to reconstruct holes of model, then we inpaint these images and do 2D to 3D transformation of new added pixels back to 3D space. This work can use any inpaint algorithms without modified. In this paper we use both structural inpainting and textural inpainting, structural inpainting is used for reconstructing depth values and textural inpainting is used for reconstructing texture of surface.