Planar surface orientation from texture spatial frequencies

摘要: Abstract This paper presents a computational model and practical algorithm for determining the three-dimensional orientation of planar surface from visual texture information. The consists three parts: (1) local spatial frequency based representation; (2) describing projection to image texture; (3) solution under an assumption homogeneity. first measures dominant at each point using wavelet-like transforms, then finds that minimizes variance frequencies' backprojections. is tested on photographs real-world surfaces, exhibiting average accuracy better than 3° in slant 4° tilt. current are more accurate, yet substantially simpler, earlier versions this approach.