Brdf Yuan Based On B-spline Fitting And Numerical Analysis

Bidirectional reflectance distribution function (BRDF) is an important tool for the field of remote sensing applications and computer graphics. The research project of××National Key Laboratory provided the backdrop for this paper, we need to fit the BRDF values that were from the actual acquisition of the project and analysis the performances of the corresponding fitting algorithms.This paper is mainly based on B spline basis function, which is the segments form of spline basis function. First, we take the project with the least-squares method of curve fitting, its running speed is fast enough, but the "over-fitting" is serious; then, we use the least squares method of surface fitting, but the running speed is too slow.B-spline function neural network is the combination between B-spline basis function and neural network, it not only retain the improvement in speed that caused by the B-spline basis function, but also inherite the advantages in terms of accuracy from neural network. This paper uses L-M algorithm to the adjustment of weight, improves its parameters initial value, and finally proves its advantages in the terms of the speed and accuracy of the algorithm through the simulation experiment.To improve the deficiencies of confidence range of the B-spline function neural network, we use the support vector machine algorithm based on B-spline kernel function. The solving of the support vector machine algorithm was converted into quadratic programming problem, while avoiding the possible problem that solution is locally in the neural network optimization problem. The simulation results show that the fitting method of support vector machine based on B-spline kernel function retained almostly the speed and accuracy of neural networks, and also overcomed the various deficiencies of B-spline function neural network.At last, the application software was realized with MFC library, and the Matlab programs that has been implemented were called in VC++6.0 through the COM component technology.