A New Robust Subspace Modeling Method
一种新的鲁棒子空间建模方法

Noise data can arbitrarily skew the solution from the desired solution in traditional subspace learning methods that based on least squares estimation,in addition,batch-based mode of subspace learning is time consuming for large scale and high dimensional problem.To deal with these two problems simultaneously,we propose a new robust learning method which comprise of two parts:robust initial parameter learning based on robust dual square function,degrade descending rule and M-estimators,which can automatically detect and remove picture level outliers and pixel level outliers;then followed by a robust incremental subspace learning process considering removing the picture level outliers and pixel level outliers.The simulation experiments show that our method is robust to different noise data,and gets better reconstruction effect of training data through learned subspace for illumination with a much more higher learning speed than Torre's method4].