It probably depends on your application: If you have a ground truth and a standard-of-reference technique for processing, then RMSE is useful for comparing the performance of your (new) approach with the standard of reference. (You can check if the RMSE of your technique is comparable or better than the RMSE of the standard of reference.)

SNR is more generally applicable since you do not necessarily need a ground truth (although this makes the evaluation simpler) or a comparison to another technique - generally, if you obtain a good SNR (>>1), then the results maybe useful; if the SNR is very low (signal comparable to or even lower than noise), then the performance is probably not sufficient ...

So, SNR can be interpreted in an absolute sense, RMSE needs to be compared to either another RMSE or to the signal (and the latter results in something similar to SNR).

One more point to consider: RMSE includes all differences relative to the ground truth, i.e. not only noise-like (incoherent) signal differences but also coherent artifacts (from e.g. translation or distortion in the case of image processing). In the case of SNR, it depends on the definition (and measurement) of the noise level - if the noise level is determined in a signal-free region, than coherent (image) artifacts may not be reflected by the calculated SNR.

thanks Olaf,

its a good description and answers most of my doubts.

Can you direct me to some literature where i can find further description regarding this matter.

No, sorry, I do not know any helpful literature discussing the applicability of these different metrics. Some of my search results (discussing at least some aspects) are attached.

http://books.google.com/books?id=YOOWBQAAQBAJ&pg=PA77

Article A Universal Image Quality Index

Article Image and Video Compression for Multimedia Engineering : Fun...