结合局部区域特征的自适应简化率网格简化算法

Mesh Simplification with Adaptive Simplified Ratio Based on Local Region Features

现有的网格简化算法通常要求人为给定模型整体简化率或者设置几何、颜色、纹理等属性的约束,如何合理地设置这些阈值对没有经验的用户来说比较困难.文中结合监督学习的方法,构建一个多层感知机模型来实现局部区域自适应简化率的预测.该感知机模型能根据网格模型不同区域的局部几何特性,提取出相应的特征,并根据这些特征进行分类.不同的分类对应着不同的简化率,从而在简化网格时根据分类对不同区域设置不同的法线偏差阈值,实现自适应简化率的网格简化算法.为证明该算法的效果,实验中选择不同种类和复杂程度的三维网格模型进行了简化,并与基于QEM能量函数的简化算法进行了整体简化率和简化后的视觉效果的实验对比,结果表明,相对于传统整体简化算法,这种基于局部区域特征设置自适应简化终止条件阈值的简化算法,能有效地根据几何特性对网格进行自适应简化,在保持模型细节的同时,提高了网格的整体简化率.

Most of existing simplification approaches simplify a mesh model by giving a global simplified ratio based on the geometry or texture properties.However,users without experience may not set a reasonable threshold.In this work,we train a multi-layer perceptron with supervised learning to predict an adaptive simplified ratio.This model classifies local regions by the extracted features from local regions based on local geometry.Each classification maps to a pre-defined simplified ratio.During the simplification process,the normal deviation threshold will be set via the simplified ratio at each local region to guide the simplification algorithm.To evaluate the performance of the proposed method,we test it on many complex mesh models and compare it with the QEM energy-based methods.The numeric of simplified ratio is computed as the evaluation rules.Experimental results show that the proposed method can effectively simplify mesh model and achieve a better global simplification ratio than traditional methods in competitive visual results.