An efficient method for specular-enhanced BTF compression

Abstract

The Bidirectional Texture Function (BTF) has been used in the data-driven model to recover the reflectance of complex surfaces. The high-dimensionality of the BTF causes huge measurement data, but that needs to be compressed when real-time rendering is required. However, the distribution of BTF data is highly non-uniform due to non-lambertian reflections. Projecting the measured data onto a linear space constructed by a statistical analysis such as PCA results in low-quality of data compression. In this paper, we propose a computationally efficient and robust data compression method. Based on the assumption that there exist many overlapping measurements, we seek to find reflectance holes caused by the non-lambertian reflection with a series of linear algorithms. Before performing the standard compression, our method separates diffuse reflections which can be well approximated with a linear model and compress them using a standard factorization. Specular components detected in the reflectance hole are separately stored using a sparse matrix representation to prevent the loss of the range of BTF data. Experimental results show that our method effectively improves the compression accuracy as well as the visual quality with a competitive computation time.