5S-1
The Design and Application of Gabor Filter Bank for Robust Normal Estimation from Texture
○劉 光耀，齋藤 豪（東工大）
We describe a robust solution of estimating surface normals from either deterministic or stochastic textures under perspective projection. A specific Gabor filter bank is designed for effectively representing the local spectral distribution and estimating the texture distortion from nearby image patches. The associated affine distortion matrix is recovered according to the pattern of the filter responses, and interpreted to infer local normal of each patch. Our approach need not explicitly identify texels or make restrictive assumptions about texture's nature such as isotropy. The accuracy of the method is evaluated on both planar and curved surfaces with synthetic textures.