Mesh Simplification with Adaptive Simplified Ratio Based on Local Region Features

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.