Why is NIQE commented out?

[choosehappy]

This is a long shot, but I was searching around the web for a python implementation of NIQE and found yours.

I noticed in the jpg_demo.py that its commented out, but there is no mention as to why.

I was wondering if its "safe" to use, or if it was commented out because it wasn't producing the correct result?

Thanks :)

[sunbinbin1991]

I use NIQE algorithm , but find it have no difference in judge whether the image is blur or not。 I don't know the reason

[aizvorski]

@choosehappy It's been a while so I don't remember exactly. I think it was producing outputs that did not match those in the original NIQE paper, although they did correlate well with the other quality metrics. The code in niqe() looks good to me, you may want to check some of the functions it depends on eg estimate_aggd_params().

Now that MathWorks has NIQE, perhaps someone who has access to that can make a basic test suite with a few example images and their NIQE values.

[aizvorski]

@sunbinbin1991 Could you post a code+images example? I think slight degrees of blur only correspond to slight changes in NIQE. You need to blur a lot to see a big change. Try size 10 or 20 gaussian blur.

[IceClear]

Hi, I just test the NIQE algorithm, using RGB images. However, the results seem totally different from those in Matlab version. Is there anything wrong? (I convert BGR to YUV and compute NIQE in Y channel, same as Matlab version)