Introduction to Pattern Recognition: A Matlab Approach

92 CHAPTER 3 Data Transformation

Observe that the projections of the data of the two classes coincide. Thus, in t hi s case Fisher’s

discrimi nant anal ysis cannot provide a smaller subspace where the class discriminati on is retained.

This happens because the number of classes is equal to 2 and so the dimensionality of the reduced

subspace is at most 1, which is not sufficient for t he current probl em.

Step 2(a). To generate a 3 ×900-dimensional matrix whose columns are data vectors, repeat the code

given in step 1(a), replacing the last two lines with

% Definition of the number of classes

c=3;

% Definition of the class label of each vector

y=[ones(1,N) ...