pythonnumpyimage-processingnorm

Python: numpy.linalg.norm produces different result than np.sum(np.abs(x)**2,axis=-1)**(1./2)

I have 2 images with size (256,256,3) and I want to calculate the norm-2 of their difference.

# Original Image

np_ori = np.array(icon_original).reshape(1,-1)

# My reconstructed image

np_0 = np.array(icon_0).reshape(1,-1)

If I use numpy.linalg.norm

norm_org_0 = np.linalg.norm(np_ori-np_0)

I get

62735

When I use np.sum(np.abs(x)*2,axis=-1)**(1./2)

np.sum(np.abs(np_ori-np_0)**2,axis=-1)**(1./2)

I get 

2207

The results are the same even if I use .reshape(-1) to turn it to vector

Shouldn't those two produce the same result?

Solution

  • As Divakar mentioned is was a datatype issue. Converting to float64 solved the problem.