Compression of View Dependent Displacement Maps
Jing Wang, Kristin Dana · 2005
Texture is important to enhance realism in surface representations for computer graphics and vision. View dependent displacement mapping (VDM) is proposed recently to give the effects of shadow, occlusion and silhouette, since traditional texture mapping can not produce all of these effects. Also, VDM can be viewed as a data structure to encode the texture geometry and exploit the modern graphics processing unit (GPU) to do interactive rendering. One drawback of the VDM approach is large memory consumption. Singular value decomposition (SVD) based compression has been proposed to alleviate this issue. In this paper , we present a method that combines k-means clustering and non-negative matrix factorization. Experimental evaluations show that the performance of our approach surpasses prior SVD-based method.