Palmprint Identification - Why do we align the images before we use the Fourier Transform?

Question

I was reading the paper PALMPRINT IDENTIFICATION BY FOURIER TRANSFORM by WENXIN LI, DAVID ZHANG and ZHUOQUN XU about identifying persons based on an image of their palm.

• The first step in their identification algorithm consists of aligning and pre-processing the image since they spent a few pages on that front, this gave me the impression, that this step is kinda important.

• What I didn't understand however: Why do we do this? The essential part in my opinion is the Fourier Transform afterwards, that is used to extract/generate the feature. But from my understanding, this should even work without this aligning/pre-processing step.

• I have found a property of the DFT that states that rotating the spatial domain contents rotates the frequency domain contents, can be read here.

• In order to extract features, some rotation might be necessary, at some point. In my opinion however, this does not justify to think up such an elaborate method of aligning/preprocessing the image, there has to be more behind that.

• Therefore, I am most likely overlooking/missing something. Thanks in advance.

Answer 1

From the article:

[...]only a sub area of a palmprint image is used in feature extraction.

The alignment by a shift and other image transformations is done in the full hand image which is then cropped to a sub image containing only the features of interest. To analyze the sub image they Fourier transform it. They are not shifting the sub image alone but the full image so the shift operation does not have a direct equivalent operation in the frequency domain of just the sub image.