Review of converted Olympus E-PM2

[enricosavazzi]

My review of the Olympus E-PM2:

 

http://savazzi.net/photography/e-pm2.html

 

It contains also a few findings of general interest:

- CMOS and CCD sensors record false-color NUV differently.

- There is yet another eBay source of converted cameras (eBay seller eeassa).

 

Season's greetings

[Andrea B.]

Enrico, thank you for the excellent review of the Oly.

 

Your website has a wealth of great information!!

 

The Ebay seller eeassa is using Spectrosil in his full spectrum conversions. This is a good choice. Although your photograph of the chipped edge, the use of silicone sealant and the gaps in the adhesive tape give me pause.

 

At this point the only Ebay seller I really would recommend is Kolari Vision.

[Damon]

Thanks Enrico. I have always been impressed by your professionalism, research, and website.

 

Since I am troglodyte when it comes to these things...

-Are you saying that if you were to take a NUV picture of a black-eyed susan for example with all 3 cameras and all those filters, the images would look different (maybe all the D70's would be similar)?

So you might not be able to pull the yellow out of the flower UV pic with the Panasonic/baader, like can be done with the D70 and baader U combo?

 

-D

[Andrea B.]

I've been able to obtain the same "final" false colour palette with Nikon, Pentax and Panasonic conversions under the Baader-U filter or Precision-U filter. Colour palette changes when using other filters, of course.

[Shane]

The use of silicone and adhesive tapes make me shudder.

 

I have seen another modifier who converts to monochrome and re-attaches the thin window (that protects the sensor) using silicone. I was unable to get a response to my question pertaining to the type of silicone used. Most silicones outgas and if this occurs in a contained environment such as the sensor cavity then the outgas products can be re-deposited onto all the available surfaces. Not good. At best it creates a cloudy coating and worse, some silicones outgas chlorine which can react with any moisture present and lead to all kinds of problems e.g. electrical leakage and corrosion.

 

Adhesive tapes also come with a similar set of issues.

 

There are "semi-conductor/electronic component" grade silicones and adhesive tapes available but are they using them and are they being applied in a dehumidified environment.

[enricosavazzi]

Just to avoid misunderstandings, the thin window protecting the chip is an integral part of the chip package and was not removed in this conversion. Only the UV-cut, IR-cut and antialiasing filter was replaced.

 

Silicone is of course problematic for the risks mentioned by Shane. In this case, so far it seems the conversion was successful. We will see if there are any long-term problems with condensation between the replacement window and the chip package window. I believe the original filter was held in a rubber gasket, and a legitimate question is why this gasket was not re-used to hold the replacement window. Nonetheless, as stated in my review, I would have spent a little more just by buying a new E-PM1 in Sweden, so the price I paid was fair even for an an amateurish but working conversion. I doubt I would have been able of doing a much better job myself, since I don't have suitable equipment to cut and grind a replacement window.

 

A properly cut replacement window of the exact size for this camera would of course have been a better choice, and I would be willing to buy such a window and do the conversion myself. Albeit not at the prices asked by LifePixel for their replacement windows: at that level, buying a second-hand mirrorless camera and removing the built-in filter can be cheaper than one of their replacement windows alone. I believe LifePixel is setting high prices for replacement windows because they don't want these sales to compete with their (good but also expensive) own conversion services. If any optical lab/company is reading this: there is a market for precision-cut replacement windows, but you must be prepared to sell at reasonable prices and to provide a broad range of sizes, shapes and thickness.

 

As for color palettes, yes, similar palettes can be obtained by tweaking in post-processing with all three cameras, especially if one starts from a raw image. My test was only made to assess the "native" performance of the sensor and processing engine (the latter, affected also by the algorithms used to determine the in-camera custom WB). So perhaps I should modify the language in my review to make this clearer. Nonetheless, I think we can state that the images straight out of the CMOS cameras show a more differentiated false-color at different wavelengths (in one case red, yellow and green, in the other case violet, yellow and green) than the images straight from the CCD camera (slightly different orange/red shades).

 

Another complication that makes Damon's question difficult to answer with a "yes" or "no" is that the images taken with different filters may contain different information even if their false-color is similar (like in the CCD camera test). By shooting multiple images and assigning each of them to one color channel of a composite image, one can generate multiple false colors even when there is no false color difference to start with.

 

As for eBay sellers of converted cameras, I saw one from Australia a few weeks ago. I did not save the link because it is not relevant to me, but might be worth a search for our Australian members.

[colinbm]

Fantastic article Enrico, thanks.

I am looking at getting an Oly EPN for UV photography next year.

 

The Sydney, Australia seller maybe this one.... http://stores.ebay.com.au/sydneylightingspecialist4life/IR-INFRARED-CAMERA-/_i.html?_fsub=4220887012&rt=nc&_pgn=1&_ipg=48

They also sell HID lamps.

They could be connected with this Sydney, Australia, camera converter ? http://www.jjcamera.com.au/

 

Cheers

Col

[Shane]

Enrico

I read your article on the review of the Olympus E-PM2 and as usual I found it interesting, helpful and useful, however I find your summary misleading.

 

“A side-by-side comparison of three converted cameras also indicates that CMOS sensors record different NUV bands as characteristically different false colors, while CCD sensors seem to lack this capability.”

 

There are many factors at work that can contribute to the different false “colours”, the least of which is CMOS vs CCD. First it is important to remember that all three of these cameras utilize luminance only sensors. Any colour imparted to the image is implemented post-capture (post-recording) and based on CFA encoded luminance data.

 

At the sensor level, Quantum Efficiency (QE) is primarily determined by the silicon junction depth, the thickness, reflectance and transmission properties of overlying sensel structures (passivation, microlenses, Colour Filter Array) and sensor structures (Internal Cut Filter/Optical Low Pass Filters, cavity cover “glass”). All these structures only affect luminance intensity and in this respect, the CFA only acts to encode the luminance data through luminance intensity differences.

 

The manufacturer uses this QE to produce a spectral sensitivity conversion profile which is implemented during conversion of the CFA encoded luminance data to an internal working colour space, typically CIEXYZ (during RAW conversion). The key to colour accuracy at this point lies in the colour matching functions (CMF), which in turn depends on accurate spectral information about the sensor.

 

CMF’s typically cover a range anywhere from 380-780nm to 400-700nm. What determines the width of the range may depend on OEM vs 3^rd party profiles and/or ICF spectral range, who knows?

 

The first significant “colour” error factor arises when comparing OEM RAW converters to 3^rd party RAW converters, as the latter do not have access to the sensor QE and must “guesstimate” based on colour chart methods similar to Xrite Colorchecker.

 

When using the OEM or 3^rd party RAW converter, problems can arise because removal of the CFA ICF, during camera conversion, invalidates the original spectral sensitivity conversion profile or 3^rd party colour (camera) profiles, particularly at the UV and IR ends of the spectrum.

 

Additional problems can arise depending at which stage white balance (WB) is implemented and how it is implemented, before or after conversion to CIEXYZ. The method is typically dependent on knowledge of the sensor spectral sensitivity and (often) proprietary WB settings (Nikon), which 3^rd party converters do not have.

 

WB implementation and/or colour matching function/camera profile, are the most likely culprit of any colour differences observed in the NUV and not CMOS vs CDD device structure.

[Andrea B.]

Shane, thank you for additional details about interpreting false colour.

We have cautioned many many times not to read too much into False Colour.

I've somewhat given up on that topic.

 

As mentioned earlier, I've always been able to "find" the standardized blue/yellow (mostly) palette that Bjørn and I use (and did NOT "invent" and never claimed to invent just to make sure one of my very picky readers gets that message) in all my Nikons, the Panasonic and the Pentax which covers a range of CCD/CMOS or whatever each manufacturer calls it.

 

It would be interesting for everyone to make use of Raw Digger for checking the actual raw data which was recorded.

[Shane]

For years I utilized dcraw and Rawnalyze (no longer available for newer models) and subsequently upgraded to RawDigger about 2 years ago. However, I have definitely not utilized its full potential as of yet but it is a great program.

[Andrea B.]

Here is an example of my poinsettia shoot using Fast Raw Viewer, a companion to Raw Digger.

 

The point to be made here is that regardless of chosen white balance, the histogram for each screen shot is displaying the underlying raw data captured -- and that does not vary.

 

I have somewhat arbitrarily chosen an Adobe Linear gamma curve to apply so that the poinsettia does not look too dark. Other gamma curves are available depending on how you wish to edit your foto.

 

I have just convinced myself to haul out the card-de-credite and pay for the demo versions of FRV and RD which I've been using. These are useful tools !!

 

The poinsettia as shot with a Nikon D610 in-camera White Balance. (Nikons will not fully WB in UV.)

 

The poinsettia with an Auto White Balance supplied by Fast Raw Viewer. Not too bad, actually.

 

The poinsettia with a Custom White Balance selected on a magenta area. This gives pretty much the standardized palette referred to above. No yellow in this scene, but a hint of blue is seen.

 

The poinsettia seen under Uni White Balance in which green is given twice the weight.

Doubling green with all the predominating red turns the poinsettia orange in this false colour scheme.

[Andrea B.]

So, it would be instructive to see the actual raw histograms for the Olympus fotos.

[Shane]

I have somewhat arbitrarily chosen an Adobe Linear gamma to apply so that the poinsettia does not look too dark.

Other gamma curves are available depending on how you wish to edit your foto.

 

I assume you are actually referring to Adobe Linear tone curve and not gamma. I believe the "Adobe Linear" refers to the linear tone curve found in ACR/LR and is in addition to any implemented working space gamma.

 

Gamma is implemented to counteract the native gamma of the capture or display devices whereas tone curves are used to compensate for visual perception. Tone curves are applied "on top" of gamma. I know that the developers of FRV are aware of that, however the limited reference material for the program under Contrast Curve seems to be a little unclear and muddy the water.

 

If you really were using a linear gamma, the image would be extremely dark in a sRGB2.2, aRGB2.2 or ProPhotoRGB 1.8, colour space. You would have to implement an sRGB1.0, aRGB1.0 or ProPhotoRGB1.0 colour space in order for it to appear "normal" brightness.

[Andrea B.]

Shane, I'm well aware of all that but I'm unsure of what the FRV folks are doing in the Tonal Curve menu because they have labeled the choices as Gamma1.8, Gamma2.2, sRGB, L*, Adobe Linear and some others. So, there seem to be a mix of gammas, colour spaces, adobe and whatnot !!! There doesn't seem to be a way to select two choices. So I chose the Adobe Linear simply because it looked the best, whatever it may actually be in this app. I will change the 'gamma' to 'curve' in the post if that helps?

 

Ideally one could see the gamma-less (dark) raw in FRV and then see the curve applied by the camera. I have not yet figured out FRV enough to produce that. Maybe I should try this in Raw Digger. Hmmmm........

[Shane]

Just noticed an error in my earlier post, now shown corrected below and in the post.

When using the OEM or 3^rd party RAW converter, problems can arise because removal of the CFA ICF, during camera conversion,

[enricosavazzi]

I modified parts of the discussion and conclusions in my review. As pointed out by Shane, several factors may be involved in generating a different false color in the different cameras, and we cannot really decide which are the most important. The observation remains that, with custom WB that produces close results in VIS, the differences in NUV false color in the different cameras are far larger than the differences in VIS color.

[Shane]

I'm unsure of what the FRV folks are doing in the Tonal Curve menu because they have labeled the choices as Gamma1.8, Gamma2.2, sRGB, L*, Adobe Linear and some others. So, there seem to be a mix of gammas, colour spaces, adobe and whatnot !!! There doesn't seem to be a way to select two choices. So I chose the Adobe Linear simply because it looked the best, whatever it may actually be in this app. I will change the 'gamma' to 'curve' in the post if that helps?

 

I came across this comment by Iliah Borg regarding the Curve menu in FRV, thought it might be useful.

 

"curves affect only the appearance of the image on the display. But for the exposure adjustment, there is a control in Preferences to switch on and off histogram shift."