What is ZCA Whitening?
Zero Component Analysis:
Making the co-variance matrix the Identity matrix is called whitening.
This will remove the first and second-order statistical structure.
ZCA transforms the data to zero means and makes the features linearly independent of each other In some image analysis applications, especially when working with images of the color and tiny type, it is frequently interesting to apply some whitening to the data before, e.g. training a classifier.