For some image SNR provides value in negative and for other image it provides positive value. Note- all image covers the same area, with same dimensions.
Hello,
Normally, SNR is positive for positive signals (no units of measurement). However, I think you are using SNR in dB (decibels), then you may get negative value as well. This is quite common for signals with less power than the noise, i.e., SNR[dB] = SignalPower[dB] - noisePower[dB], if power of the noise is higher than the signal then SNR < 0.
The following article may be of use to you.
A New Image Restoration Performance Measure with High Precision
Steven S. O. CHOY, Yuk-Hee CHAN and Wan-Chi SIU
Link: http://www.eie.polyu.edu.hk/~enyhchan/c_97oe.pdf
the SNR improvement may yield measurements of large spread when it is applied to a series of homogeneous restorations. This indicates that the SNR improvement is of low precision.
SNR improvement contains only one reference point in it, namely zero, which is to indicate `no improvement in SNR'. Thus a positive SNR improvement indicates that the image fidelity is improved, while a negative one indicates deterioration. However, the
minimum of SNR improvement is at (-infinity)and it happens when there is no distortion in y (y = x). It does not correspond to `the maximally achievable deterioration'. Furthermore, the SNR improvement will encounter problems in interpreting the amount of fidelity improvement when one of the following three cases happen
SNR improvement can not act as a good restoration performance measure, for that
it convey unclear and, sometimes, even wrong message on the image fidelity improvement in a restoration.
Hence apply a weighted sum of the pixel fidelity improvement as a measure of image restoration performance.
P.S.
SNR is appropriate and was designed for one-dimensional signals, such as detection-only radars.
Later, as these systems evolved, one had to include the effects of interference. This gave rise to the SNI (Signal to Interference Ratio).
In image analysis, I introduced a 2-D metric of Target to Interference Ratio (TIR) to measure how easy or how difficult it is to analyze, detect, or recognize objects in images.
I think I have included a report of mine in Research Gate, titled: Classifiability of Infrared Images. I will check that it is there. If so, you can go to my profile to get the reference.
- Badges
- Science topic
- Similar topics
- Computer Science
- Images