Smooth Locality Sensitive Discriminant Analysis Method
Chunming Xu · Jisuanji gongcheng · 2011
Locality Sensitive Discriminant Analysis(LSDA) is a recent proposed supervised feature extraction algorithm.LSDA can not only utilize the class information but also consider the the intrinsic geometrical structure of the data.However,LSDA is a vector based method so it neglecteds the spatial correlation of the pixels in the image,and its performance may be degraded in this case.To solve the problem,this paper proposes a Smooth LSDA(S-LSDA) method.It introduces a spatially smooth regularization which incorporates the spatial correlation information into the objective function of LSDA.It shows that the derived coefficients are spatially smooth and the extracted features are more effective for classification.Experimental results on face image databases show the effectiveness of the proposed algorithm.