The current metrics are MSE, MAE,SNR, PSNR, cross correlation etc Low values of MSE,MAE and high values of SNR, PSNR and cross correlation close to 1 indicate good denoising. But if you consider a signal say x(t) which is noisy and another signal y(n) which is very similar to x(t) and no noise is removed, then the values metrics appear to be such that the signal denoising is good. Infact if you consider simple smoothing filters, the lesser the smoothing, the lesser is the noise removed but "better" are the values of metrics. However in this scenario, no to very less noise has been removed. Thus the metrics are in essence not giving us valuable information. Would like to hear your views about this.
Dear Aditya
in this manner, I think some structural based and frequency based metrics may be useful and give valuable information. two of most frequent of them list below:
Relative energy (RE) : The RE metric simply compares the ratio of the total energy of two time signals.
frequency-dependent normalized root-mean-square (NRMS).
I hope be useful.
regards
Dear Ali,
I can understand the usefulness of frequency dependent NRMS but how does RE help? How does signal energy correlate to noise removal?
Dear Aditya
visual quality metrics (MSSIM, FSIM, PSNR-HVS-M, PSNR-HMA) do a much better job at quantifying denoising image quality than conventional MSE, PSNR and other metrix.
Dear Aditya
I found to you some useful new and good denosing metrics , please follow the next links:
http://www.cs.cityu.edu.hk/~qiyang/publications/iccv-13/
http://alumni.soe.ucsc.edu/~xzhu/doc/metricq.html
Please do not hesitate in any inquiry
Thanks
Dear Aditya,
Please refer the attached paper. It will give you a recent technique for specific assessment of image denoising quality metric.
Conference Paper A new model for performance evaluation of denoising algorith...
Dear Ruslan and thank you for your interesting answers.
Dr. Mohammed thank you I shall look into those links.Thanks a lot shall get back to if I require further clarifications
Mr. Vikrant thank you I shall look into your paper. In fact Mr.Vikrant I had come across your paper of morphological filtering (top-hat based) of signals and was looking into finding the optimal sizes of the filtering element for the same. I shall look into your publication also. Thank you
Regards
Aditya Sundar
feature similarity index (FSIM),
structural similarity (SSIM),
visual information fidelity in pixel domain (VIFP)
these metrics can help, their source codes are published online.
References:
FSIM: L. Zhang, D. Zhang, and X. Mou, FSIM: A feature similarity index for image quality assessment, IEEE Trans Image Process 20 (2011), 2378–2386.
VIFP: H.R. Sheikh and A.C. Bovik, Image information and visual quality, IEEE Trans Image Process 15 (2006), 430–444.
SSIM: Z. Wang, A.C. Bovik, H.R. Sheikh, and E.P. Simoncelli, Image quality assessment: From error visibility to structural similarity, IEEE Trans Image Process 13 (2004), 600–612.
Apart from the full reference quality measures such as PSNR, MSSIM, UQI etc, you can use blind (non-reference) quality measures such as BRISQUE and NIQE and others by the authors of these measures.
http://live.ece.utexas.edu/research/quality/nrqa.htm
This could be a nice metric
https://www.sciencedirect.com/science/article/abs/pii/S1746809421005590
- Badges
- Science topic
- Similar topics
- Geoscience
- Atmospheric Sciences
- Atmosphere