Phash vs SIFT when identifying a similar image

Question

I have questions about all sift and phash

First of all, I use SIFT to identify a similar image in real time. Like photos on a camera phone, a small amount of rotation and a blurry effect can be.

And I found Phash . So, I am testing phash on my demo page. But the result made me breathe.

This is the test result above:

In this test, two images are captured along the x axis. So t have a rotation. But the logo of the correct images was removed, and the person was moved to the left side. In my eye it is "very similar." In addition, SIFT will completely catch it.

Now, that is the question.

• Is pHash faster than SIFT?
• How reliable is pHash accuracy?
• SIFT output is too large for real-time use. Therefore, I have to use a hash to create a smaller output size, such as LSH (location sensitivity) . Any other way to try?

Answer 1

OK I understood.

pHash cannot recognize rotation and critical motion as one.

In the case of the information space, pHash was very useful to use. This is a very small size: one image for one hash. SIFT, however, needs 128 bytes to get the function. And in one image there are many elements.

In the end, SIFT can identify a similar image, not pHash. But more and more was needed.

In the speed scanner, I still can not check. But I think that pHash was faster than SIFT, because SIFT should work for many functions in one image.

If you have other answers to the above question, please tell me.