OJODIGITAL

since the beginning i had problems with my G1 which showed labyrinth artefacts mainly in the left bottom corner of the image with most raw converters. especially EAHD in raw therapee showed very ugly artefacts.

first panasonic service didn't even want to exchange my G1 for a new model, as they didn't see this problem as a defect, but i insited on it, showing them image-examples from other G1 that didn't have this problem at all or only slightly.
so i finally got a new G1 but this one shows nearly the same effects as the old one.

someone suggested to use global green equilibration to solve this issue, here is the original thread: Re: G1- strange sensor mosaic problem - experts please have a look!: Panasonic Talk Forum: Digital Photography Review

unfortunately when i tried i don't manage to see any difference in perfect raw regarless if i turn green equilibration on or off. does this only work with particular demosaicing algorithms. i tried VCD and AFD so far.

thanks for any help!

best regards, olaf

Olaf, stop worring and love your G1!

Panasonic sensors (inside Panasonic and Olympus cameras) are a bit different from others, as they have differente wavelenghts in their green captors.

I think it is a design decision in order to make their cameras more sensitive to green differences, but in fact it does not let you develop your images with some great interpolation algorithms that needs to work with three (R, G and B) channels instead of four (R, G1, G2 and B).

As far as I know, you only have a four channel option in dcraw (when using -f and thus forcing VNG interpolation), choosing VNG-4 in RawTherappe or using both bilinear-4 and VNG-4 in perfectRAW.

Unfortunatly, VNG is not the best interpolation algorithm out there, and so, you would like to use something more advanced like AHD, EAHD (in RT), AFD, etc.

When I got at that point I investigated myself the reason behind the interpolation artifacts until I got the answer: if you reequilibrate the green channels before interpolation, then you can use any interpolation algorithm you want. That was in fact the origin of perfectRAW development.

First I started equilibrating the green channels mean value, as I though that the channels response should be quite linear. After some tests it got clear that something more locally-adaptative was been needed. The response was local green channel equlibration (after a global one).

So, you must choose local green channel equilibration (do not worry about the tile radius, it is OK as default) and any interpolation algorithm that you want.

I have realised that Adobe's ACR is doing something similar, as they mark their DNGs with a "need green equilibration" mark when needed (although it is very poor documented in their DNG SDK).

In the near future you will be able to choose more interpolation algoritms in perfectRAW (EAHD, HPHD like in RT maybe) but also a new interpolation algorithm that is being developed by Emil Martinec with excellent results (far better that any other you have already seen! ).

Take into account that the perfectRAW GUI your are seing is provisional as we are developing an OpenGL/wxWidgets GUI cross platform (it will run natively in Linux, Windows, and MacOS in 32 and 64 bits) for the 1.0 version. We are also planning to port most of the development code into the GPU so it will also be lighting fast (that is for 1.5 version).

Our improvements for high ISO images are also better than those of DxO Optics. perfectRAW is designed to get you the most possible information (detail for example) from your RAWs. I hope you enjoy using perfectRAW with your G1 images!

Best regards,

__________________
Manuel Llorens

Olympus E-510, E-300
Mis fotos

manuel, thanks a lot for your response.

i love my G1, but i was concerned that it might have some defect as most raw converters gave me those artefacts. of course i know about the different RGGB filter from olympus cameras. but there you always get these artefacts evenly over the whole image. with my G1 only parts of the image were affected.
when i compared my G1 raws to other raws, i found that others didn't show this problem that seriously. for example if you look at the G1 raw-file from dpreview, there are hardly any artefacts visible:
http://www.dpreview.com/reviews/Pana...W/P1060154.zip

if you look at my examples, you will see that the left bottom corner shows quite a lot of them. i still think that it is a kind of inaccuracy of color-filter or anti aliasing filter that affects some models more than others.

i uploaded 2 raw-files here if you want to have a look. as they are in portrait format, you have to look at the upper right corner:
zusatzbeispiele - eSnips, share anything

so during my search the only raw converter that worked flawlessly was capture one, but it doesn't officially support the G1, so you have to patch the raw-files first.
all other converters including ACR also produced those artefacts with my G1, although not that much as raw therapee for example. now that gabor also has implemented your green equilibration method it shouldn't be a problem, but so far it is only active for olympus raws, and you don't have the option of turning it on or off.

are there other problems that might occur when using green equilibration?
i quickly tested one of my files with prefect raw and local green equilibration as you suggested, and the labyrinth artefacts are indeed completely gone, but i found some areas that started to show new effects, have a look at the pylon. with green equilibration there is some ghost-halo effect visible to the left (you can find this raw file is also in the previous link):


i am really interested in trying out new demosaicing algorithms, because none of them really produced clean results so far.
but i discovered a new algorithm from a polish guy "jacek". he calls it "DCB" and it produces stunning results. it extracts slightly less details than HPHD for example, but diagonals are butter-smooth without any stairsteps. and you can sharpen up the images really a lot without introducing any artefacts. zipper noise is non existent with DCB. have a look at his site:
LinuxPhoto.org

i asked jacek to convert some of my images and the results were nearly perfect. have a look here:


i already contacted gabor and asked him to implement DCB into raw therapee, and although he agreed, he didn't confirm any release-date. so it could still take a year to get it done. i know jacek is a very open guy. maybe you could also try contacting him and asking about implementation into perfect raw.
the problem is, that i don't have any raw converter that gives me good results with my raw-files. raw therapee gives nice details but too much artefacts. ACR is neither sharp nor artefact-free and it corrects all lens-distortion from G1. jacek's DCB would be an option, but converting RAWs without GUI like with dcraw is not really a funny task.

what about emil's algorithm, do you have some examples to look at? you really made me curious!

thanks again and best regards, o.

The channel diference depends on image colors and white balance. If you develop the linked image with no white balance (multipliers = 1 1 1 1) you will see it is very green. Maybe that is the reason it does not show many artifacts. I do not think that your camera is different from that.

ACR is using AHD and so it will render any image noise as line artifacts and labyrinths. Well, we think that the user must have the control of the development process, not the software author. In perfectRAW you can choose wheater use green channel equilibration or not. Gábor approach is very different: he prefers an easy to use product.

Ok... I see now . I will limit green channel equilibration when the channels difference is too big, that should get rid of those artifacts . I have made some tests. Better than that is to limit the pixel different before taking into account one pixel in the tile.

As in many interpolation algorithms when you use DCB your are getting smoother diagonals at the price of lossing important detail or missing the interpolation direction:
(i will post some samples soon)

I do not see such and improvement with EAHD. Emil's algorithm will be much better, sure!

Well, it must be Emil who decides when his algorithm is ready for the world. Be patient!

Best regards,

__________________
Manuel Llorens

Olympus E-510, E-300
Mis fotos

of course there is some price to pay. until now i always wanted maximum details, but these are results i am getting (ACR):


meanwhile i reached such a point that i would rather accept slightly less details if there were no artefacts.

everything depends a lot on the camera. i discovered that G1 is quite a "hard" one. there is not a single demosaicing algorithm that is able to produce both fine details and smooth edges, without any zipper-noise or stairstepping. so far DCB really convinced me. as a side effect it also produces quite a low base noise level.

if Emil's new algorithm is better, it won't say no of course.

so i am looking to seeing some of your samples!

bye, o.

oluv, I have done some tests with your images and I have concluded that the green channels difference is highly correlated with the image content. So it does not look as a sensor fail.

It may be somehow greater in the borders of the image, probably related with the lens vignetting.

I am redoing the green channels equilibration in perfectRAW to get rid of any possible artifact. I will post samples as soon as done, but I do not have much time at this moment.

Best regards,

__________________
Manuel Llorens

Olympus E-510, E-300
Mis fotos

Esta vez voy a contestar en español antes de que me tiren de las orejas. Dado que las imágenes hablan por sí mismas creo que no le costará a oluv seguir el texto.

Finalmente he rehecho el equilibrado de canales verdes. Es un poco más rápido y potente que antes. Además no crea artefactos y se puede controlar, en vez del tamaño de la celda como antes, lo que son laberintos y lo que es textura. De ese modo podemos arriesgar más o menos, si bien he ajustado los parámetros para que por defecto quite todos los laberintos de mi Oly E-300 y de la Panasonic G1 de oluv sin crear demasiados artefactos.

Yo creo que los resultados son buenos, los laberintos desaparecen del todo y los artefactos son mínimos. De hecho, si uno ve solo la imagen a 1x sin laberintos no creo que identifique ningún artefacto.

En cuanto optimice un poco más subiré una nueva versión.

Aquí tenéis el mismo ejemplo de oluv. Sin equilibrar verdes, equilibrando con el nuevo algoritmo, la misma imagen enfocada y el cielo de la misma imagen sin y con equilibrado de verdes:
__________________________________

(Google translation, no se entiende ni papa )

This time I will answer in Spanish before I pull the ears. Because the images speak for themselves I do not think that will cost oluv follow the text.

I finally remade balancing green channels. It's a little faster and more powerful than before. Moreover, it creates artifacts and can be controlled, rather than the cell size as before, which are mazes and what is texture. Thus we can put more or less, although I have adjusted the parameters to the default remove all the labyrinths of my Oly E-300 and the Panasonic G1 to oluv without creating too many artifacts.

I think the results are good, they disappear from the labyrinths and all the artifacts are minimal. In fact if one sees only the image at 1x without identifying labyrinths do not think any artifact.

As optimize a bit more upload a new version.

Here is an example of oluv. Without green balance, balancing the new algorithm, the focused image and the sky of the same image with and without balanced green:
__________________________________

All images, 2x zoom, AHD interpolation, 2 passes median filter B/R channels with perfectRAW 0.66:

Without green eq.:

With new green eq.:

New green eq., sharpened (sorry, I forgot colorspace conversion ):

The same image, sky without green eq.:

And with the new green eq.:


And 1x AFD post-processed image:


Un saludo:

__________________
Manuel Llorens

Olympus E-510, E-300
Mis fotos

thanks manuel for your demonstration and sorry for not having responded for so long.
i had a lot to do lately. in the meantime i was also testing several demosaicing algorithms as i am still unhappy with the results i am getting from my G1.

which is going to be the new algorithm in perfect raw you mentioned before?
i did some comparisons with several algorithms and although DCB is softer than all the others, it produces really nice edges, which is perfect for architectural photography. it doesn't do well with nature though, because it tends to create patterns that start to look artificial.

i wanted to show you some of my tests, of course i can also upload the original raw-files for testing, but G1 RAWs seem to be quite hard for most of the available demosaicing algorithms.
i sharpened all equally, except DCB which was sharpened stronger, because the original output is softer than the other algorithms.



if you look at the wheel cover from the mercedes, you will see that none of the algorithms manages to render it well, only DCB shows hardly any artefacts here.
HPHD is worst, but also shows most details.
RGE does quite well with the number plates, but all other except DCB struggle with sharp edges.

i am not sure if i don't prefer the softer and less detailed output that DCB produces. sure, it is less detailed, but in many cases all algorithms produce too much artefacts, especially with diagonal edges.

here is another example which i did with an older version of DCB. all were sharpened equally. the only algorithms that do well in this case are DCB and the one from Capture One: