normalization , colums, rows

Hi everyone,

May I know which way correct to do normalization for a matrix by colums or rows?

Thanks.

[Ameer Hamza]

The correct way depends on what you are trying to do, how are you normalizing etc. You need to specify what is your purpose for normalizing the matrix.

[Dhurgham Al-karawi]

Thanks for your reply. I wanna do classification using SVM. The dimension of the matrix is 242* 256.

Answer 1

[Jan]

x = rand(242, 256),
xNRow = x ./ sum(x, 2);   % Auto-expanding, Matlab >= R2016b
xNCol = x ./ sum(x, 1);

For older Matlab versions:

xNRow = bsxfun(@rdivide, x, sum(x, 2));
xNCol = bsxfun(@rdivide, x, sum(x, 1));

Now the rows or columns are normalized, such the the sum is 1.0. But perhaps you want the norm to be 1.0?

xNRow = x ./ vecnorm(x, 2);   % Auto-expanding, vecnorm needs >= R2017b
xNCol = x ./ vecnorm(x, 1);

Or with older Matlab versions:

xNRow = x ./ sqrt(sum(x .* x), 2));   % Auto-expanding, >= R2016b
xNCol = x ./ sqrt(sum(x .* x), 1));

or again with bsxfun.

There are more methods for a "normalization": Set the mean to zero, and/or the std to 1 or such that the maximum peak height is 1.0. So you have to find out, what you need mathematically. Then the implementation in Matlab is easy.

[Dhurgham Al-karawi]

Thanks for kind information. It seems that the normaliztion is done for both colums or rows. Is it okey to do normaliztion for colums only.Because when i do it for rows only the performace is getting low but with colums it is very high. That's why am asking which one correct rwos or colums. Thanks again

[Dhurgham Al-karawi]

I have used the following formula to do normalaztion

minData=min(min(Class1_feature))

maxData=max(max(Class1_feature));

Class1_feature=((Class1_feature-minData)./(maxData));