When and why do we need to perform Fourier transform before calculating the phase of the image?
The discrete Fourier transform (DFT) does by definition transform your image into the frequency domain, mapping it to a matrix of complex numbers.
Again, by definition, a complex number has a phase and a magnitude component.
Thus, you don't "need to perform FT before calculating the phase of an image", you get the phase of the image as a result of the FT.
It's a bit like saying "why do I need to perform an addition before calculating the sum of 2 number?".
I would suggest you to learn more about Fourier Transform