E-M5 has DR 13!

Started Apr 13, 2012 | Discussions thread
Detail Man
Forum ProPosts: 15,000
Like?
Clarification about how Digital Camera DR and SNR are specified
In reply to Detail Man, Apr 14, 2012

Detail Man wrote:

Sergey Borachev wrote:

Can the technical experts who have been bombarding pixels with photon accelerators and dissecting noise particles emitted comment on what the Techradar review's verdict regarding E-M5's stupendous image results?

I would really like to hear from those who have previously analysed to great details RAW files comment about the review finding that it has a DR of 13 , and that it is the best of all compact camera. Is that believable?

TIFF images (after conversion from raw) from the OMD have a better signal to noise ratio than those from the Fujifilm X Pro 1, Panasonic GX1, Sony NEX-7 and Olympus E-P3, showing that the camera copes well with noise .

Regarding: http://cdn1.mos.techradar.com//art/cameras/LabCharts/Olympus/Olympus_EM5_TIFF_SNR-580-100.JPG

Note that the same (it appears un-named) RAW converter show a Signal/Noise Ratio of around 41.0 dB. A linear ratio of 112.20. Take the logarithm to the base 2 and you get 6.81 EV ("stops").

Measured Dynamic Range cannot in any case exceed 3 dB (41.4%) greater than the measured Signal/Noise Ratio - which the same web-page reports as being 6.81 EV ... 13 EV would be 36.50 times greater. Kind of makes one wonder what is going on "upstairs" in their voluminous heads ...

TIFF images (after conversion from raw) have a high dynamic range across the sensitivity range, with the OM-D achieving the highest result we've seen for any compact system camera .

Regarding: http://cdn1.mos.techradar.com//art/cameras/LabCharts/Olympus/Olympus_EM5_TIFF_DR-580-100.JPG

I have yet to discover the "Higgs Boson", but it seems that you may well have discovered some real "Techno Bozos". But, since they are whispering "sweet nothings" into the ear ... go for it !

This remains yet to be determined. I will let it stand for now until DxOMark publishes test results ...

Note: Just remember that 12-bit analog/digital converters have less than 12 EV Dynamic Range

(Perhaps) the E-M5 does indeed use a fancy non-linear compression scheme in a manner similar to the Sony NEX-7. Time will tell, and dreamers have more time to entertain sweet dreams. However, (if so), the E-M5 circulating about these days would be compressed (needing the inverse process of digital expansion to result in a process-able RAW image-file) - and that would have been readily apparent to any/all persons who have obtained those image-files. No indication of that to date ...
.

My statements about DR not being able to exceed SNR by more than 3 dB related to the case (as in specifying audio signal processing devices) where the maximum linearly reproduce-able signal-level is divided by the root-mean-square (RMS, vector-summed) value of the "noise-floor" (composed of random as well as periodic noise sources as amplified at the point in the signal-chain of measurment). In those cases of measuring DR and SNR, the peak value of the maximum linearly reproduce-able signal-level is used in the calculation of DR, and the RMS value of the maximum linearly reproduce-able signal-level is used in the calculation of SNR. Since the ratio of the peak to the RMS-average of a sinusoidal test-signal is the square-root of 2 (1.414, or 3 dB when expressed in units of voltage or current), this is why I said what I said (about maximum possible DR related to measured SNR by a difference of 3 dB). I neglected to use my only average brain before posting.

In the case of a digital camera, DR is measured by taking the ratio of the image-sensor system's maximum (so deemed linear, perhaps ambitiously, by somebody ... ;)) output signal when illuminated , divided by the RMS value of the image-sensor system's "noise-floor" when not illuminated - or, in the case of DxOMark, apparently very slightly illuminated to the point where the ratio of the signal corresponding to that slight illumination is equal to the RMS value of the "noise-floor" that exists in that particular situation. Thus, when the Signal/Noise Ratio equals 1.0 (0.0 dB).

In the case of a digital camera, (the maximum) SNR is measured by taking the ratio of the image-sensor system's maximum (so deemed linear, perhaps ambitiously, by somebody ... ;)) output signal when illuminated divided by the RMS (vector-summed) value of the sum of all noise-sources (Read Noise, Dark Noise, Photon Shot Noise, etc.). As a result of the existence (almost entirely) of Photon Shot Noise in that situation, the SNR is indeed smaller in value than the DR (as can be seen in DxOMark's RAW-level test-results). I write this in order to amend my misguiding of the readers ...
.

That being said, I do think that TechRadar's test-results seem perhaps just "a bit" (actually more like "two bits" or so) hard to believe. I don't see any mention of the RAW converter/processor used. That in itself seems rather odd, and lacking in reasonable specificity. Please provide a reference if you yourself have found such information regarding test procedures utilized. Dream on!

DM ...

Reply   Reply with quote   Complain
Post (hide subjects)Posted by
Keyboard shortcuts:
FForum PPrevious NNext WNext unread UUpvote SSubscribe RReply QQuote BBookmark post MMy threads
Color scheme? Blue / Yellow