基于极限投影变换的3D模型检索

3D Model Retrieval Based on Terminal Projection Transformation

基于认知理论和投影理论,提出极限投影面积方法对三维模型进行坐标标准化:将模型不断绕过模型重心的基向量旋转,采集每次坐标面投影面积,在多次迭代后得到最大面积投影,用同样办法获得另一坐标面上的最小面积投影,建立模型的坐标系.实验结果表明:该方法标准化的坐标系处理对象范围宽,适用于网格模型、点云模型和各类曲面模型,针对模型噪声、简化、攻击等有很强的鲁棒性.经该方法标准化的三维模型在3个坐标面的投影作为模型的特征描述子,并在实验中为实验模型库建立了对应的特征描述子库,使得对模型检索转化为特征描述子间的比较检索.检索实验表明:该方法针对增加了特征描述子的模型库具有较快的检索速度;同时具有较强的鲁棒性,但检索的精度稍差.

Based on the related theories of projection and cognition, this paper introduces a way to standardize the coordinates of a 3D model. During the continuous rotation of the 3D model around a vector passing through its center, the direction along which the model has the maximum projection area, can be found and defined as the Z-axis. Then by rotating the 3D model around the Z-axis, one of its orthogonal directions along which the model has its minimum projection area, is defined as the Y-axis. In this way, we can define a standardized coordinate system of the model, which is not only invariant to the object location and orientation but also robust to noise, model simplification and attacks. Our method is applicable to a wide range of objects, such as mesh models, cloud points models and other surface models. The projections of the 3D model on the three coordinate planes can be regarded as a feature descriptor for 3D model retrieval. Experimental results show that our method has a fast retrieval performance with strong robustness, but it is slightly accurate.