http://pro.sony.com/bbsccms/assets/files/cat/camsec/solutions/E_CMOS_Sensor_WP_110427.pdf
See page 5. It seems that they are doing further noise reduction in hardware after they convert to digital. Verrrrrrry interesting! There might be more to the theory that the chip is doing this before the imaging pipeline even sees the data. I'd kill for a white paper now. On Fri, Nov 29, 2013 at 10:35 AM, Zos Xavius <zosxav...@gmail.com> wrote: > Larry, > > If what falconeyes is suggesting is correct, it looks like the > firmware is actually manipulating the data. From what I understand, > these chips quickly read out all their raw data to the imaging > processor which interprets the data and applies whatever they have > cooked up in the firmware to massage it. This means erasing hot > pixels, some noise control, etc. I'm pretty sure the imaging sensor is > a mostly dumb chip, though I do know that there usually some basic > hardware noise control built into the chip. I think, if I am reading > this correctly, is that falconeyes theory is that it is taking > information from the green channel (the most sensitive channel) and > interpolating some of that information into the other channels to > smooth luminance noise. It seems to be mostly accepted fact that the > K-5 does up to iso 1600 in hardware and that after that it seems to be > pushing 1600 in software and dealing with the noise in the imaging > pipeline. Yeah, a white sheet on EXMOR imaging chips would be pretty > revealing I think. If you ask me. I think looking at the loss in > resolution in the RGB channels would be the most revealing as to what > they are doing to the data. If I were a betting man, I'd throw a lot > of money on the idea that all manufacturers manipulate their raw data, > especially at high ISO. I think it would be foolish to assume > otherwise. Some people think that the K-5 is native up to 12,800. > There might be some truth to that, but I don't think it changes that > it seems to be manipulating the raw data. > > On Thu, Nov 28, 2013 at 3:25 PM, Larry Colen <l...@red4est.com> wrote: >> On Thu, Nov 28, 2013 at 01:44:02PM -0500, Zos Xavius wrote: >>> See also: >>> >>> http://www.pentaxforums.com/forums/115-pentax-k-5/118892-how-iso-3200-works-your-k-5-technical-2.html >>> >>> Especially falconeye's comments. >> >> Very interesting. There are times like this when I seriously wish that >> I could peek at the source code. I wonder if the data sheets for the Sony >> Exmoor sensors are available? >> >> I suppose that one possibility would be to take a 16 stop grey scale, >> photograph it at different ISOs and exposures and compare the data in >> the raw files. >> >> As a geek, I want to know every detail about how the process works, >> where data is being thrown a way. As a photographer, what I really >> need to know is "what gives me the best results in which circumstances?". >> As a geek, I like to think that understanding the process would let me >> figure things out from first principles. As a lazy photographer, >> I just set the ISO and do the best I can with the light available. >> >> Assuming that high ISO is done with math rather than electronics, it >> is an interesting question as to whether a smaller processor closer to >> the sensor can do a better job than a bigger processor on a desktop >> computer. >> >>> >>> On Thu, Nov 28, 2013 at 1:38 PM, Zos Xavius <zosxav...@gmail.com> wrote: >>> > Boris, >>> > >>> > http://www.imaging-resource.com/PRODS/K5/K5RAW.HTM >>> > >>> > "Here, we can see the Pentax K-5 clearly produces the cleanest looking >>> > RAW files, though it appears to be applying some noise reduction at >>> > higher ISOs (above ISO 1,600), which cannot be turned off. The noise >>> > reduction applied is pretty subtle, though, nothing like the >>> > heavy-handed approach used in earlier Sony SLRs. Still, it's something >>> > we'd rather not see in RAW files, as it does impact fine detail." >>> > >>> > This is noted by many other people as well. The noise reduction >>> > settings are for JPEG only by the way. The only affect how much NR is >>> > applied to a jpeg if your are converting raw or shooting in JPEG. >>> > >>> > On Thu, Nov 28, 2013 at 2:00 AM, Boris Liberman <bori...@gmail.com> wrote: >>> >> Zos, two points that still keep me wondering: >>> >> >>> >> 1. I couldn't find any mention of RAW noise reduction in Imaging >>> >> resource review after this link: >>> >> http://www.imaging-resource.com/PRODS/K5/K5A.HTM >>> >> >>> >> 2. I see what the person on DPReview wrote, but I think there is a >>> >> confusion and (as explained in imaging resource review) the noise >>> >> reduction can be configured per ISO. I don't remember exactly, but it >>> >> stands to reason that ISO 3200 is where it is set to kick in by >>> >> default. >>> >> >>> >> Specifically, once I learned about this setting, I went and configured >>> >> both of my cameras to apply no noise reduction until ISO 12800 or so. >>> >> >>> >> So, I'm yet unconvinced that K5 applies noise reduction at ISO 3200 >>> >> just because. >>> >> >>> >> On Wed, Nov 27, 2013 at 7:24 PM, Zos Xavius <zosxav...@gmail.com> wrote: >>> >>> Sorry for the delayed reply Boris. >>> >>> >>> >>> http://www.dpreview.com/forums/post/36834206 >>> >>> >>> >>> Also read the Imaging Resource and dxomarks reviews of the K-5. Pretty >>> >>> much all the reviews noticed the RAW NR above 1600. Given the K-5's >>> >>> output at 3200, Pentax made a good choice IMO. Their hardware NR is >>> >>> pretty decent, very clean looking, and still retains a lot of detail, >>> >>> especially after the raws are processed gently in lightroom. I never >>> >>> once objected to the engineers' decisions on the K-5 when it came to >>> >>> the sensor. Pentax seems to be very good at massaging maximum >>> >>> performance out of a sensor. The K-3 IMO looks very good too. A 50% >>> >>> increase in density with similar noise and DR over the previous >>> >>> generation is very good IMO. The next round of chips will likely be >>> >>> even better. What's not to like? The K-5 I have is very good in IQ. >>> >>> I'll keep shooting it till it doesn't take pictures anymore. >>> >>> >>> >>> On Tue, Nov 26, 2013 at 7:46 AM, Boris Liberman <bori...@gmail.com> >>> >>> wrote: >>> >>>> The source of information that indicates that K-5 applies smoothing to >>> >>>> RAW files even if I specifically set its settings not to do so. >>> >>>> >>> >>>> On Tue, Nov 26, 2013 at 1:52 PM, Zos Xavius <zosxav...@gmail.com> >>> >>>> wrote: >>> >>>>> The source for K-5 raw smoothing? The source for resized K3 files? >>> >>>>> What source do you speak of Mr Boris? :P >>> >>>>> >>> >>>>> On Mon, Nov 25, 2013 at 11:45 PM, Boris Liberman <bori...@gmail.com> >>> >>>>> wrote: >>> >>>>>> On 11/26/2013 6:38 AM, Zos Xavius wrote: >>> >>>>>>> >>> >>>>>>> Also I must add that the k-5 does some definite noise reduction to >>> >>>>>>> the >>> >>>>>>> raw files over iso 1600, whether that's in the imaging pipeline or >>> >>>>>>> the >>> >>>>>>> firmware is neither hear nor there at this point. At 3200 on the k-5 >>> >>>>>>> you are most certainly trading resolution for image quality. I don't >>> >>>>>>> have any real figures, but I feel that looking at the files from my >>> >>>>>>> k-5, its easily giving up quite a bit of resolution due to noise >>> >>>>>>> smoothing. From what I can gather the K-3 is resolving more fine >>> >>>>>>> detail but yet showing similar amounts of noise when resized to k-5 >>> >>>>>>> resolution. That's what I have seen with my own eyes at least. To be >>> >>>>>>> honest they are pretty close and careful raw processing will result >>> >>>>>>> with more detail in the k-3 files with similar noise levels. I find >>> >>>>>>> pentax's jpeg engine to be not the greatest at noise reduction, but >>> >>>>>>> honestly, its better than some other camera makers too. >>> >>>>>> >>> >>>>>> >>> >>>>>> Zos, can you please point me to the source of that information? I'm >>> >>>>>> extremely interested to *know*. >>> >>>>>> >>> >>>>>> Thanks. >>> >>>>>> >>> >>>>>> >>> >>>>>> >>> >>>>>> -- >>> >>>>>> PDML Pentax-Discuss Mail List >>> >>>>>> PDML@pdml.net >>> >>>>>> http://pdml.net/mailman/listinfo/pdml_pdml.net >>> >>>>>> to UNSUBSCRIBE from the PDML, please visit the link directly above >>> >>>>>> and >>> >>>>>> follow the directions. >>> >>>>> >>> >>>>> -- >>> >>>>> PDML Pentax-Discuss Mail List >>> >>>>> PDML@pdml.net >>> >>>>> http://pdml.net/mailman/listinfo/pdml_pdml.net >>> >>>>> to UNSUBSCRIBE from the PDML, please visit the link directly above >>> >>>>> and follow the directions. >>> >>>> >>> >>>> >>> >>>> >>> >>>> -- >>> >>>> Boris >>> >>>> >>> >>>> -- >>> >>>> PDML Pentax-Discuss Mail List >>> >>>> PDML@pdml.net >>> >>>> http://pdml.net/mailman/listinfo/pdml_pdml.net >>> >>>> to UNSUBSCRIBE from the PDML, please visit the link directly above and >>> >>>> follow the directions. >>> >>> >>> >>> -- >>> >>> PDML Pentax-Discuss Mail List >>> >>> PDML@pdml.net >>> >>> http://pdml.net/mailman/listinfo/pdml_pdml.net >>> >>> to UNSUBSCRIBE from the PDML, please visit the link directly above and >>> >>> follow the directions. >>> >> >>> >> >>> >> >>> >> -- >>> >> Boris >>> >> >>> >> -- >>> >> PDML Pentax-Discuss Mail List >>> >> PDML@pdml.net >>> >> http://pdml.net/mailman/listinfo/pdml_pdml.net >>> >> to UNSUBSCRIBE from the PDML, please visit the link directly above and >>> >> follow the directions. >>> >>> -- >>> PDML Pentax-Discuss Mail List >>> PDML@pdml.net >>> http://pdml.net/mailman/listinfo/pdml_pdml.net >>> to UNSUBSCRIBE from the PDML, please visit the link directly above and >>> follow the directions. >> >> -- >> Larry Colen l...@red4est.com http://red4est.com/lrc >> >> >> -- >> PDML Pentax-Discuss Mail List >> PDML@pdml.net >> http://pdml.net/mailman/listinfo/pdml_pdml.net >> to UNSUBSCRIBE from the PDML, please visit the link directly above and >> follow the directions. -- PDML Pentax-Discuss Mail List PDML@pdml.net http://pdml.net/mailman/listinfo/pdml_pdml.net to UNSUBSCRIBE from the PDML, please visit the link directly above and follow the directions.
