How to use PLCC metrics to evaluate algorithms whose outputs are lower the better the image quality is?

[Zhang-Jing-Xuan]

Some evaluation algorithms, such as Brisque, have a lower result, the better the image quality, in this case how to calculate the PLCC to compare with those algorithms that have a higher result the better the image quality? If they use the same human eye score, then Brisque will always be negatively correlated with the human eye score, and the calculated PLCC will always be negative.

[chaofengc]

You can simply use the absolute PLCC value if the metric is lower better.

[Zhang-Jing-Xuan]

Thank you for your reply. Can I take the inverse of the results of algorithms (x->1/x) where the smaller the result the higher the quality and then calculate the PLCC?

[chaofengc]

It is OK to do so, but I think not a good practice. Because PLCC measures the linear relationship, but 1/x is not a linear function. You can simply normalize the values with min and max value like $1 - (x - min)/(max - min)$

[Zhang-Jing-Xuan]

OK. Thank you for your reply.