Zero Mean Normalization

Zero-mean normalization is also called Z-score normalization. The values of attribute X are normalized using the mean and standard deviation of X. A new value new_v is obtained using the following expression:

new_v = (v - Ux)/Sx 

where Ux and Sx are the mean and standard deviation of attribute X, respectively. If Ux and Sx are not known they can be estimated from the sample (column in D that corresponds to feature X). 

In such a case, we can substitute Ux1 for Ux and Sx1 for Sx in the expression above.
After zero-mean normalizing each feature will have a mean value of 0.

 Also, the unit of each value will be the number of (estimated) standard deviations away from the (estimated) mean. 

Note that z-score normalization may be sensitive to small values of Sx.

When zero-mean normalization is applied, all the data in each profile are slid vertically so that their average is zero:

The blue profile was obtained from the red one by a vertical translation so that the average of the profile be zero.

From：http://blog.163.com/penansord@126/blog/static/74642408200931541934535/