Can anyone suggest me the standard interpretation values (Tables) for PSNR (Peak to Signal Noise Ratio) and Kappa coefficient (Accuracy Assessment parameters) for Image Processing to check whether the value generated as the outcome of the proposed module falls within the acceptable range or not?
@ 2D Image @Tabular representation @ Intervals
Hi Rebika,
I have used PSNR quite extensively to check for the amount of distortions introduced to images (quality measurement), but I am not aware of any standard table in conjunction with image processing. I, however, came across several research papers that mention an acceptable PSNR level of 40 dB or above.
Dear Rebika Rai,
PSNR is used to compare two images, it is straight forward that it is a similarity measure. It is defined in terms of MSE. As Abbas Chedded tols, a PSNR of 40dB reflects a more closer set of images. PSNR is used in Image compression, denoising, deblurring, enhancement to some extent. PSNR is used to quantify the level of smilarity or agreement between two datasets. Coming to Kappa coefficient, kappa does not quantifies the level of agreement between two datasets. It represents the level of agreement of two datasets corrected by chance. I think here n image processing, there no such scheme works by chance, and kappa has nothing to do. Off course, that depends on the scheme you use and the task you perform.
The range of PSNR values depends strongly on "the kind of degradation or the problem that you address" and the "image that you use to compute the value". For example, restoration results from low-level denoising is likely to have higher PSNR values, whereas the ones from more complex distortions might result in low PSNR values. Similarly, a flat image will have very high PSNR value, irrespective of the degradations, whereas the PSNR of an information rich image will have strong dependency on the kind of degradation.
Moral story: For the problem that you tackle, identify a standard dataset where researchers have reported their PSNR values and compare your value with existing works to know "whether the value you got make sense".
- Badges
- Science topic