11 Replies Latest reply on Jul 9, 2011 11:04 AM by Hudechrome

    More Usable Image Detail - Adobe Camera Raw Beats Canon DPP

    Noel Carboni Level 7

      Let me start with a question:

       

      When you do your raw conversions, do you convert to an image that's got the same number of pixels as your camera's imager has photosites?

       

      If so, why?  I ask because I'm convinced that - with modern software - to get the most usable detail out of a raw image one should convert directly to an upsampled size.  You might think that in trying to get "more megapixels from the same camera" I'm just fooling myself, but again and again when I look at the problem of maximizing the output from existing equipment (which, face it, is everyone's goal, no matter how good the equipment) I keep seeing that there is an advantage to selecting from the upsampled resolutions in Camera Raw.  Of course, the computer must be up to handling the additional data.

       

      In actual terms, my Canon 40D has 3888 x 2592 photosites (plus some guard pixels around the edges) making 10 megapixels, yet I find converting to 6144 x 4096 (25 megapixels) to be advantageous - ESPECIALLY when I use my sharpest lenses.

       

      I happened to be doing a head to head comparison between Canon 50mm f/1.8 and Canon 50mm f/1.4 lenses, and the difference between the two appeared pretty subtle - UNTIL I converted to the aforementioned upsampled resolution, and then the more expensive f/1.4 lens clearly stepped ahead.  There was simply extra detail I hadn't been seeing clearly at the native pixel size of the camera's imager, but which stood out clearly at the larger size.

       

      This reaffirmed my previous observations that conversion directly to upsampled resolutions in Camera Raw brings out additional detail.

       

      This screen grab shows part of the converted image, as produced by DPP and Camera Raw, both at 6144 x 4096 pixels as produced by the converters and displayed at 100% zoom.  Note that the detail (e.g., in the terrain) is finer and more real looking in the Camera Raw conversion, while the DPP image seems to have more sharpening artifacts.

       

      DPP_vs_ACR_Conversion.jpg

       

      If you're interested in experimenting to see how much detail you can lift out of this same image, the raw file is here:

       

      http://Noel.ProDigitalSoftware.com/temp/IMG_5400.zip

       

      I don't have a copy of any other converters, for example Capture 1, to see how they'd do.

       

      -Noel

        • 1. Re: More Usable Image Detail - Adobe Camera Raw Beats Canon DPP
          Jeff Schewe Level 5

          I would guess it's due to the adaptive upsampling in Camera Raw/Lightroom and the potential for better capture sharpening...

          • 2. Re: More Usable Image Detail - Adobe Camera Raw Beats Canon DPP
            Noel Carboni Level 7

            Can you please define "adaptive upsampling" a little better?

             

            Whatever's going on, it's pretty clear it's doing a better, finer job than DPP (or upsampling with a bicubic process from the "native" sized data).

             

            -Noel

            • 3. Re: More Usable Image Detail - Adobe Camera Raw Beats Canon DPP
              Jeff Schewe Level 5

              Noel Carboni wrote:

               

              Can you please define "adaptive upsampling" a little better?

               

              Unlike Photoshop that 3 separate flavors of resampling; Bicubic, Bicubic Smoother and Bicubic Sharper, ACR/LR can interpolate between the 3 flavors...it'll use simple Bicubic for small upsampling and blend into Bicubic Smoother as the upsampling increases. Same deal for Bicubic Sharper when downsampling...

               

              It's subtle but it does make ACR/LR incrementally better than Photoshop for resampling that can't interpolate between the 3 flavors.

              1 person found this helpful
              • 4. Re: More Usable Image Detail - Adobe Camera Raw Beats Canon DPP
                Noel Carboni Level 7

                Thanks.  Now that you mention it I recall you saying something about that before.

                 

                Just to clarify, for a given image input and output size, is Camera Raw using different algorithms on a pixel by pixel basis?

                 

                -Noel

                • 5. Re: More Usable Image Detail - Adobe Camera Raw Beats Canon DPP
                  Noel Carboni Level 7

                  I think you're right about the upsampling being special in Camera Raw, Jeff.

                   

                  I've been doing some experimentation with the fractal upsampling in Perfect Resize 7, starting with a "native sized" conversion from Camera Raw, and I've been able to equal the detail level in a Camera Raw conversion directly to an upsampled resolution, though at a cost of quite a bit more CPU time.  An advantage gained is that jagged lines are reduced and edges are cleaner with Perfect Resize, as expected.

                   

                  But none of it would be possible without the super detail that Camera Raw is extracting from the raw data.

                   

                  -Noel

                  • 6. Re: More Usable Image Detail - Adobe Camera Raw Beats Canon DPP
                    Hudechrome Level 2

                    My, oh my! Tons of subjectivity here.

                     

                    I noodled the concept and came up with this:

                     

                    First, I am not interested in the differences between converters, at least for the moment. So I took an image which was processed through DXO optical corrections and duplicated it. It is in the dng format. Next, I opened one in ACR, made sure no additional optical corrections are applied, and sent it to PS at the native 12.1Mpx value of the D90. Next, I opened the dupe,  upped the sampling to 25Mpx and sent it to PS.

                     

                    In order to compare apples to apples, we must take into consideration the magnification chosen to compare. If you set both to 200% the objects in the frame will be of different size, and we are not comparing it correctly. So the 25Mpx is left at 200% while the 12Mpx I set at 287% by trial and error so that switching back and forth there is no jump in size of the component being examined.

                     

                    Under these circumstances, the 25mpx image appeared sharper than the 12 mpx, but did not contain any more detail. What I saw is the local contrast at the transition edge was better,an what appears to be sharper was actually the onset of artifact. Mind you, none of this would, at normal print sizes, viewed without any visual aid, single out one print as better than the other. And, in normal post processing, I would push a bit past the DXO actually at ACR sharpening values about the same as the default values without any noise reduction. If I run the output at 25Mpx, I would not need or want that step. But I also have a 143M file vs 69M file!

                     

                    One final check was to take the 143M file and reduce it to 69M so that the pixel dimensions matched, using Bicubic. Now, all differences disappeared. I could not tell them apart.

                     

                    The subject was tiny branches of an ivy plant growing up a stucco wall. It was hand held at 1/100sec at 85mm zoom length using a VR lens. I am going to try it again, this time with the Nikon 50mm f1.8, on a tripod. Unfortunately, I won't use the same subject as the ivy is now fully leafed out and likely not going to show much of the undergrowth.

                     

                    I'm not sure how much subjectivity I actually removed, if any!

                     

                    I can do a converter comparison, that is DXO vs ACR  as well. But it is unclear to me whether I am actually looking at the conversion by DXO or by ACR. No tone or color adjustments were used in DXO, and they claim that the lens corrections are sidecar information appended to the original. However, noise reduction is applied to the pixels, so it would seem that conversion has taken place. If so, I'll have to run the test again, sans DXO noise reduction.

                     

                    It may be moot though, because all I am looking at here is 12 vs 25Mpx output differences.

                     

                    Message was edited by: Hudechrome

                     

                    Message was edited by: Hudechrome

                    • 7. Re: More Usable Image Detail - Adobe Camera Raw Beats Canon DPP
                      Jeff Schewe Level 5

                      Considering DxO outputs a demosiaced processed file, Camera Raw's demosiacing is not used when opening that DNG. As a result, the combination of ACR demosaicing, noise reduction and sharpening plus upsampling–the ACR processing package if you will–isn't being used. So, what Noel is testings isn't what you tested.

                      • 8. Re: More Usable Image Detail - Adobe Camera Raw Beats Canon DPP
                        Hudechrome Level 2

                        Thanks Jeff.

                         

                        I didn't think so either, but I wasn't sure of the details. I will run the test again on the basis of using the .nef alone. I hope that Noel will offer results that compare the Canon RAW in ACR with both resolutions as well.

                         

                        Nonetheless, for my workflow, it is valid as I do run this way on a regular basis and for my workflow, upsampling out of ACR shows little value.

                         

                        I have the option to upsample as well in DXO, so I can circumvent opening it in ACR and run it directly into PS as a tiff. This should level the playing field a bit.

                        • 9. Re: More Usable Image Detail - Adobe Camera Raw Beats Canon DPP
                          Noel Carboni Level 7

                          If an oddball combination of converter logic from several sources were to yield the "best" looking result (and by "best" I also mean the best input for subsequent editing/processing operations) then I would consider adopting such a combination.  The difficulty in doing so would have to be a factor, however.

                           

                          In this case, in the past I had shunned DPP's ability to produce slightly higher detail over ACR, simply because of the ease and convenience of opening files directly into Photoshop through ACR.  You can imagine I'm all the happier now that the technique I use - with the 2010 process - happens to produce the best results I have been able to get.  As I mentioned, though, I haven't tried all the alternatives.  There are just so many hours in the day!!

                           

                          What all of this DOES show is that every time there's an advancement in converter software then one's workflow for producing images from raw files needs another look.

                           

                          And I see no reason to think that converter technology has topped-out.  That means that there could be an even newer process in the future - or another converter presently available - with which even better results can be obtained from the very same files.  Imagine an algorithm involving true deconvolution, for example, that with today's CPUs might take 5 minutes to complete, but which would take only seconds on a GPU...

                           

                          On another note, I keep wondering whether I should move up to a newer model camera, and frankly (with these incremental improvements in the converter helping me make better images than ever with the old camera) I just can't justify it.  That means more money available for software!  Adobe has succeeded here.

                           

                          -Noel

                          • 10. Re: More Usable Image Detail - Adobe Camera Raw Beats Canon DPP
                            Noel Carboni Level 7

                            Hudechrome wrote:

                             

                            I hope that Noel will offer results that compare the Canon RAW in ACR with both resolutions as well.

                             

                            Not sure I follow you completely...  Just to be clear, can you lay out what "both resolutions" means?  I'll be more than happy to go through different processes and present comparison images - I just want to be sure I understand just what you're asking for.

                             

                            Oh, and you're right - some subjectivity necessarily will appear here.  I might just like the "look" of one process while you prefer the "look" of another.

                             

                            Lastly, one of my workflows involves sharpening images with my own fractal sharpening actions, and I am testing to see which converter process output produces the best result from that as well.

                             

                            -Noel

                            • 11. Re: More Usable Image Detail - Adobe Camera Raw Beats Canon DPP
                              Hudechrome Level 2

                              Noel Carboni wrote:

                               

                              Hudechrome wrote:

                               

                              I hope that Noel will offer results that compare the Canon RAW in ACR with both resolutions as well.

                               

                              Not sure I follow you completely...  Just to be clear, can you lay out what "both resolutions" means?  I'll be more than happy to go through different processes and present comparison images - I just want to be sure I understand just what you're asking for.

                               

                              Oh, and you're right - some subjectivity necessarily will appear here.  I might just like the "look" of one process while you prefer the "look" of another.

                               

                              Lastly, one of my workflows involves sharpening images with my own fractal sharpening actions, and I am testing to see which converter process output produces the best result from that as well.

                               

                              -Noel

                              I had to read that twice to make sure I knew what it meant!

                               

                              What I am trying to say is to show the results of both converters in the native resolution and the 25M upsampled.resolution.

                               

                              On another note, I have been playing around with the details of noise and sharpening in DXO and ACR, looking at 300%. It's both good and bad. If you are willing to mess with the ACR settings at 300%, then ACR has the edge, even comparing to manual tweaks in DXO. They do correctly optimize, however, except that point isn't as good as ACR. The depressing point is all the tweaking to do on an individual basis. Can you imagine tweaking 100 or more images that way? Then you get to move over to other corrections.

                               

                              We are splitting tiny hairs here, and the bottom line is if you are going to make 30x40 prints from a given file, you may want to use ACR all the way and bite the bullet. If the Auto corrections for the lens employed in ACR is well done, as it seems for my Nikon 18 to 105, the workload is relaxed.

                               

                              Finally, the ACR corrected image will need final sharpening at higher values than the DXO, and at that point, they come together a bit more...sharpening anomalys and all that.

                               

                              The final fuss levels takes me back to my 8x10 days, where these kinds of determinations (max quality level) takes place under the focusing cloth, and deliberate selection of POV is carefully considered. Today, it's in front of the monitor. But instead of a few sheets of film to process, we have 100's to do.

                               

                              I thoroughly enjoy working with the files on the computer, but not so much having to deal with less than optimum technical details right out of the camera. Barrel distortion? Hasselblad was not satisfied with the "normal" corrections in it's general optics, which were damn good, so came out with the 100mm Planar that was awesome! All they gave up was max aperture, and picked the FL at a value that provided for the best corrections. So I recall anyway.

                               

                              Why should we be required to do these kinds of post corrections? Digital editing involves throwing away information, and ther appears to be enough so that correcting these deficiencies still gives a wonderful image. Just imagine what it would be like if all you need to do is to open the file and adjust values to taste, with all that information density available exclusively for that part. You don't even need to use a focusing magnifier!

                               

                              Gotta run!

                               

                              Lawrence