Sketch-based 3D Shape Retrieval

In the past, people can only obtain 3D data through 3D laser scanners, or manual editing in 3DMax and other modeling software. In recent years, the rapid development of 3D data capture technology, whether it is based on 3D scene reconstruction of images and videos, or Intel’s depth camera and Microsoft’s Kinect, make it cheap and easy to capture 3D data of models and scenes. And these technologies are going to diffuse into personal consumer products, such that smartphones, tablets and laptops will soon install depth cameras. Predictably, the number of 3D models will boom. So if we can retrieve these models, it becomes equally important and useful as text retrieval and image retrieval.Similar to image retrieval, there are several user interfaces for 3D model retrieval. But interfaces based on sketch are the most natural and convenient. In this paper, we first choose directions and approaches to project models on 2D plane. Second, we extract feature descriptors for the views and sketches, which are compact, robust and easy for indexing. At last, we match input sketch with models based on framework of bag of features or metric learning, and rank the results according to relevancy.In this paper, we introduce above stages in detail, and implement a sketch-based 3D model retrieval system. We implement the user interactive and display interface with Qt, and use OpenMesh and OpenGL to read and display models, every modules of the system are coding in C++.