Page 147 - Algorithm Collections for Digital Signal Processing Applications using MATLAB

136 Chapter 4

Step 2: Reshape the matrix into the vector whose elements are collected
column by column from the matrix.

Step 3: Co-variance matrix of the collected vectors is computed. Eigen
values of the co-variance matrix and the Eigen vectors
corresponding to the significant Eigen vectors are computed (In this
example 6 Eigen vectors are computed)

Step 4: Eigen vectors are reshaped into the matrix of the original size. They
are called Eigen ears as given in the figure 4-2

Step 5: The Eigen ears are orthogonal to each other. They can be made
orthonormal to each other by normalizing the vectors.

Step 6: For every ear image matrix, feature vectors are obtained as the inner
product of eigen basis vectors and the reshaped ear image matrix .

Step 7: Mean vector of the feature vectors collected from the same person is
treated as the template assigned to that corresponding person. This is
repeated for other persons also. Thus one template is assigned to
every person and they are stored in the database.

Step 8: To classify the unknown ear image as one among the four cate-
gories, template is computed as the inner product of Eigen basis
vectors (Eigen ears) with reshaped normalized unknown ear image.
The template thus obtained is compared with the group of templates
stored in the database using Euclidean distance.

Step 9: Template corresponding to the minimum Euclidean distance is
selected and the person corresponding to that template is declared as
the identified person.