Linear discriminant analysis is a way of feature extraction and dimension reduction. It is widely applied in face recognition

speech recognition

and handwriting recognition etc. However

many linear discriminant analyses are "hard" ones and every data point belongs to one class or another class strictly. In this paper

a fuzzy uncorrelated discriminant transformation (FUDT) is proposed based on uncorrelated discriminant transformation (UDT). FUDT is a supervised learning method with fuzzy set and its discriminant vectors satisfy the equation of generalized rayleigh quotient. Furthermore

the projection of samples to fuzzy uncorrelated optimal discriminant vectors is uncorrelated by FUDT. The experimental results show that FUDT is better than UDT in extracting the feature of SAR images which come from MSTAR.