PyTorch | Preprocessing images

Table of Contents:

• Three kind of preprocessing
• Normalization
• Standardization
• Whitening images

Article not finished…

Three kind of preprocessing

These are:

• normalization
• standardization
• and whitening

Normalization is for the mean, standardization is for the variance and whitening is for the correlation.

Normalization

…

Standardization

…

Whitening images

The most interesting whitening algorithm is called Zero Component Analysis or ZCA.

Whitening also called sphering means we will transform the image so that we get the covariance matrix is the identity matrix.

Identity matrix means all one in the diagonal and 0 else.

Happens to be the white noise has the same covariance matrix as the image after whitening.

Here are more details on the identity matrix.

Whitening is a bit more complicated than the other preprocessing, but we now have all the tools that we need to do it. It involves the following steps:

Zero-center the data
Decorrelate the data
Rescale the data