FAST OPTIMAL LINEAR APPROXIMATION OF IMAGE RECOGNITION OF VARIABLY ILLUMINATED SOLID OBJECT

摘要

PROBLEM TO BE SOLVED: To perform the efficient calculation of a low-dimensional linear partial space optimally including the group of pictures to be generated by changing illumination made incident to the surface of a three-dimensional object at a plurality of different relative position of the three-dimensional object and an observation camera. SOLUTION: The matrix elements of a spatial covariance matrix corresponding to an object are calculated for the arbitrary preliminarily decided distribution of an illumination condition. The arbitrary pair of a normal vector and albedo is approximated by the set of all the pairs of albedo and normal having the center of a cluster being the result of the vector quantization of the set so that the maximum complication can be reduced for the model. Then, a covariance matrix which is not dependent of a complicate but practical view point is prepared for the object, and made diagonal off-line. The covariance matrix which is dependent of the view point is calculated from the result of diagonalization which is not dependent of the view point, and made diagonal on-line in a real time.