If you could completely remove the noise in a signal, I'm sure you would win the Nobel Prize.
Since no one has won the Nobel Prize for noise removal yet, you're stuck with imperfect noise removal methods. There are thousands of them with new ones coming out every month. You have simple ones like box filter averaging, and median filter, to more complicated ones like bilateral filter, sigma filters, mean shift filter, etc., to more complicated and more effective methods like BM3D, non-local means, K-LLD, K-SVD, UINTA, etc. etc.
The bottom line is how effective does your noise removal method need to be to measure the object you want to measure to the accuracy that you require? Maybe the denoising method is not great and you can measure something to within 1% of the true number, but maybe that's fine because if you're within +/- 5% you're able to tell if this part you're inspecting passed or failed.
