In the previous chapter, we described a feature selection method based on a linear transformation of the feature space. A recognition system is constructed by adding a classifier at the latter stage. In contrast, the subspace methodSubspace method introduced in this chapter is an interesting method that uses a linear transformation of the feature space itself for classification without separating feature selection and classification. The history of this method can be traced back to the 1960s, when Watanabe Watanabe, Satoshi et al. noted that when many feature vectors are plotted in a multidimensional feature space, they are often unevenly distributed in a subspace Subspace of very small dimension in the feature space. By utilizing this property of uneven distribution of feature vectors, it is possible to focus only on the subspace where data is distributed when performing classification.