Polygon Subdivision Control Using SVM With CSF

Abstract

In 3D computer graphics, less is better if reducing the number of polygons in a model has no visual impact. In this paper, assuming a method that controls polygon subdivision level to reduce the rendering computational cost, we propose a method that uses a Support Vector Machine (SVM) to choose one from two subdivision levels by the perceptibility difference between the adjacent level of detail. The SVM takes four features obtained in the rendering procedure as input and performs a binary classification to determine whether a polygon should be divided or not. Our experimental results show that the trained SVM performs a binary classification with 72% accuracy.