Talk:Peak signal-to-noise ratio

How is this different from regular SNR? — Omegatron 03:06, 1 December 2005 (UTC)

Omegatron, by regular SNR I suppose you mean (average signal power)/(average noise power). That is usefully calculable for signals that have constant average power, such as AM or FM radio signals (audio has no DC component). Video is an example of a signal that does not have constant average power but has constant peak power.Cuddlyable3 07:17, 25 May 2007 (UTC)

Variants
Similar to the SSIM page, I think the PSNR page should include information on variants such as PSNR-HVS and PSNR-HVS-M. There's not a ton of info on the web that explains these nicely so it'd be nice to have. I myself don't really have a great understanding on it so I can't write the part but I'm suggesting it as an addition.

PSNR-HVS: Peak Signal-to-Noise Ratio taking into account Contrast Sensitivity Function (CSF),

PSNR-HVS-M: Peak Signal-to-Noise Ratio taking into account Contrast Sensitivity Function (CSF) and between-coefficient contrast masking of DCT basis functions.

Sources for research: http://www.ponomarenko.info/psnrhvsm.htm Jiecut (talk) 18:38, 12 March 2016 (UTC)

simplify
So is higher or lower PSNR better? Can someone add this to the article? Daniel.Cardenas 23:09, 7 August 2007 (UTC)


 * A higher PSNR would be better, to further extend on this a PSNR of 100 would be exactly identical to the original. 71.183.91.117 02:54, 16 October 2007 (UTC)

Ahem... Why would PSNR 100 be identity? It is just a very good signal, but noisy nontheless. If you mean identical for all practical purposes, I would agree, but choosing 100 suggests that this is a special value like 100%, which it is not. -- 06:38, 7 November 2007 (UTC)

A higher PSNR is indeed better. To see this don't take what one says as granted but rather try to understand the underlying equations, they are not difficult. The MSE is basically the difference between the original and reconstructed image pixels, if the image is compared to itself (perfect recontruction or identical) then the MSE will be 0 as the difference between all pixels and their counterparts in the reconstruction will be zero. Now in the PSNR calculation we are dividing by the MSE, now a MSE of zero gives us an undefined PSNR, however if you think of it in terms of approaching zero rather than zero itself then the PSNR increases as the MSE approaches zero. That is the smaller the number on the denominator of the division, the larger the result. Thus I would have to say that a PSNR of 100 is far from correct for an identical image, a PSNR of infinity (or undefined to be mathematically correct) defines an image identical to the original. Publications on wireless image transmission quality define PSNR values of 20 dB or 25 dB for adequate image quality . I would say 30 dB to 40 dB are of quite good quality from my own work but this was for a uni assignment so you are to take it with a pinch of salt until you try it yourself. Shall I add this description to the main article? Keep in mind I'm still a newbie at Wikipedia. Theclem54 (talk) 04:47, 3 June 2008 (UTC)

ac vs dc peak power
Hello. For "one-sided" signals such as Video/Luma I can see that the definition matches the equations; but for AC signals (such as Audio, which is signed, but also arguably Video/Chroma which is kinda unsigned) the definition of "peak energy" seems off by a factor of 2. Could some knowledgable author/editor clarify this please? Thanks. 212.44.20.129 (talk) 16:55, 19 June 2009 (UTC)

Confusion regarding the value of MAX used in PSNR calculation
This page says that the value of MAX is (2^B -1)

i have read (http://www.mathworks.com/matlabcentral/fileexchange/135-psnr) that the value of MAX is the max difference between the pixels of the two images.

will the MAX for 16 bit images be 65535?

my images being 48 bit (3*16) I used 1/(m*n*3) to calculate MSE.

I am trying to convert an RGB image to YUV and convert the YUV back to RGB. so there is some loss introduced but not as much as a lossy compression. using Matlab's rgb2ycbcr and ycbcr2rgb functions, i was expecting a high PSNR value.

I was surprised to find that the PSNR was 100.1818, the MSE was 0.4118

but if I use 'max difference' as MAX i am getting 75.84 which seems to be more correct.

if use set MAX to 255+255/256 as specified in http://compression.ru/video/quality_measure/info_en.html#psnr I get 52.01

please help me (ramprasad85 at gmail.com)

Ramprasad N (talk) 08:39, 11 August 2010 (UTC)

User, 17th Dec. 2020: there is a minus missing in the last line of the PSNR equation!(equation image)  — Preceding unsigned comment added by 94.212.190.11 (talk) 09:46, 17 December 2020 (UTC)


 * It sounds like you have a colour image (3*16bit), rather than a high-depth greyscale (1*48bit). This makes a big difference! If a pixel value is 0xff0000, you should probably interpret that as 255/3 (mean intensity across colour channels), or something similar - not 16711680 (the 48 bit number, given in decimal). Your max range would then be 256, not 16777216. But don't take my word for it, I'm not an image processor, just thinking aloud ;-) GyroMagician (talk) 09:22, 11 August 2010 (UTC)

Original uncompressed image is much less sharp then compressed...
is there a reason for this? Is compression operating a hidden sharpening ? If so it should be mentioned, if not the original may be wrong. — Preceding unsigned comment added by Massimociacci (talk • contribs) 14:09, 26 October 2014 (UTC)
 * Something somewhere is doing a very poor job of filtering the thumbnail that you see on the page. The actual original in a lossless PNG file (click through to see), and all compressed files were derived from that original.  Yes, it's annoying and confusing. Weedwhacker128 (talk) 21:54, 26 October 2014 (UTC)