Upsizing you 10 megapixels to 30 mp sort of...

Status
Not open for further replies.

benjikan

TPF Noob!
Joined
Feb 8, 2007
Messages
454
Reaction score
14
Location
Paris, France
Can others edit my Photos
Photos NOT OK to edit
Pro Applications..Making 10.2 megapixels in to 30.6 mega pixels, sort of...
There are ways to improve the rendition of the K10D that will give it similar resolution to a Pro Back of 31.6 megpaixels. I will be shortly doing an Ad Campaign with the K10D where I could have opted for a Digital Back. I will shoot in RAW and introduce three layers with a microscopic shift on each of the layers. After doing so I will sharpen the layers at between 0.3 and 0.5 pixels at 240 to 300 percent. I "Free Transform" the two duplicate layers by "1" Pixel each and I do not "Flatten" the image. It works quite well. What I do is use two colour layers and one B&W layer. I still use the "Soft Light" setting. I am effectively getting the impression of a 30.6 megapixel image. The pixels are all slightly offset. It really does work quite well.

One more pointer.

You take the original for example, I don't know let's say 2400 x 3600 as the original layer. The first duplicate I would make 2399 x 3599 and the second 2401x3601 and when applying Unsharp Mask I would sharpen at 0.3 to 0.5 pixels at 240 to 300 percent to compensate for the offset. Do not Flatten the layers as that would negate the size of the file.

Thought I would share that with you. Theoretically it is not 31.6 mega pixels, but it sure looks like it.

Here is a sample using "Free Transform" to offset the image "1" pixel per layer. What you are seeing is a crop of about 66 percent from the original that can be viewed on one of my sites under the series name "Divine Inspiration".

Ben
Sample Here:
http://www.stevesforums.com/forums/view_topic.php?id=568680&forum_id=78&jump_to=730336#p730336

From Here:

http://anashcreation.com/thenashgallery/BenjaminKanarek/chr_tiencloseup?full=1
 
I'm sorry but that doesn't make any sense. In fact, there's now way you could mathematically conclude that you have come close to tripling the number of pixels via the process you outlined. For that matter, it also stands to reason that you can't make a 10mp image "look like" a 30mp image. It either is or isn't. And I'm willing to bet that if you took your new image and actually put it next to a 30mp image of the same subject, you would notice quite a quality difference. The only way to artificially increase the resolution of an image is through pixel interpolation, which your method does not do.

Your process is neither mathematically nor logically sound. For whatever reason, you seem to enjoy making excuses for why you shoot with a K10D. I'm sure it's a very nice camera. I own several Pentax cameras in different formats and love all of them. But unless you were actually foolish enough to turn down the offer to shoot with a proper digital back, then to be honest I think that you're lying, because the only cameras that MF digital backs are made for are at the very least much better suited to fashion photography, and in most cases they are better cameras on the whole.

If you think that this actually works, then more power to you, but I think that you're kidding yourself. You know as well as I do that everyone who is anyone in the fashion industry worldwide uses MF (usually digital). I'm not saying that you're a nobody, but I am saying that I think your post is rather silly. If you actually have the power to purchase and use a real MF digital camera, but choose not to for whatever reason, then that's perfectly fine. But I think it defies all reasonable explanation why someone who is so in love with their camera that they claim to have turned down the use of a real camera (in the fashion industry), would attempt some process that seeks to mimic those other cameras.
 
LMAO. You just basically called BenjiKan, a professional fashion photographer, a fashion photo n00b! LMAO!!!!!
 
I'm curious about this as well. I'm a total amateur, but the way I see it, you're not increasing the pixel dimension of the image. You're simply layering pixels upon pixels in a vertical manner. If you started with a 2400x3600 image, your output is still 2400x3600. You mentioned not flattening the image because you lose the "size of the file", and it seems to me, that's about all you'd lose. That is, if you were to print the image, your printer wouldn't care that you've got three layers in photoshop.

Perhaps the problem is that I don't understand what you mean by "looks like" 30 megapixels. Looks like it where? Does it look better for large prints perhaps? Does it look on screen what a 30MP image would look like a 33%? (not sure why that would be beneficial)

Honestly interested to learn more about what you're doing.
 
you're not increasing the pixel dimension of the image. You're simply layering pixels upon pixels in a vertical manner

Not quite...it sounds like he is layering them, but offset from each other by sub-pixel amounts. Also, it shounds like one technique is to use pure luma (brightness, b&w) for one of the offsets.

So a possible way to look at it:

Stack an image 3 times, offset them by sub pixel amounts from each other, blend them (pick a blend mode), then sharpen.

I'm making no comment on the effectiveness of the technique, I'm just stating what I think is happening. I'd be willing to bet that the offsetting+blending works to offset any small gradient harshness, while the sharpening seeks to reduce the blending on edges.

Also, think about it...the sensor on your camera isn't pixel per pixel accurate (unless you have a Foveon sensor)...you have the Bayer arangement of sensors (RGBG), which the camera firmware has to convert to actual pixels. So, even a 10 mega-pixel camera is not truely such...it's that the firmware tries its hardest to look like one.
 
If you offset by one pixel three times, then you've gained an extra three pixels on the x-axis and three on the y-axis. You are also forgetting that the extra pixels are only appearing at the periphery of the image.

So let's do the math.
x-axis: 3600px wide*3px=10800px
y-axis: 2400px tall*3px=7200px
= a total of 18000px added (we won't worry about the overlap for now)

Mega=1x10^6
Thus, 18000/(1x10^6)= 0.018mp

Congratulations! You've added 0.018mp to your image.

Where did you put so many pixels? At the periphery, where nobody will notice them anyway.

You haven't added any quality to your image because you haven't interpolated the pixels. You are confusing file size (in mb) with image quality. Your "final" image is very large in photoshop because you have three 2400px by 3600px images layered on top of each other, not because you've interpolated any of the pixels. It is not possible to detectably increase the resolution of an image without pixel interpolation. Therefore, it is not possible to make a photo of any given size "look" like it is of a higher resolution. In fact, if you were to flatten your three-layered image, you would find that the file size correlated almost exactly to the addition of 0.018mp.

LMAO. You just basically called BenjiKan, a professional fashion photographer, a fashion photo n00b! LMAO!!!!!

I'm not doubting Benjikan's talent, I have looked at his work extensively and I think it is very good.

I am saying two things:
1) This process he invented is complete rubbish.
2) I don't believe that he would find it necessary to try to invent such a thing if he legitimately believed that his K10D was as good as the digital MF cameras that are standard in the fashion industry. Therefore, to his claim that he turned down the use of one of those cameras, I conclude that he's either foolish or lying.

Step to that.
 
If you offset by one pixel three times, then you've gained an extra three pixels on the x-axis and three on the y-axis.

Congratulations! You've added 0.018mp to your image.

I believe you are wrong here...he is not offsetting by 1 pixel, he is (attempting) to offset by 1/2 a pixel. What that gains you is debatable, but it IS a bit more complex than just shifting everything over by 1 pixel. I don't know any better whay to explain it. Again, a 10 Megapixel Bayer interpolated image is not "really" 10 Megapixel, but the effects of the firmware make for an "interestingly hi-res approximation". His method may be a way to interpret the data a bit differently. Is her "really" getting the same image as a 30 Megapixel non-Bayer interpolated image. No. Is he getting an image that some may consider "as good as" a 30 Megapixel image once it's in print? I don't know. Sometimes "faked" looks better than "real" due to the human visual system.

It also depends on how the photos will be used. Will they be blown up to 40 feet tall for an outdoor add campaign?

It is not possible to detectably increase the resolution of an image without pixel interpolation.

Interpolation seems to be exactly what he is going after. By attempting (notice the word ATTEMPTING) to shift each image plane by 1/2 a pixel, and then blending, he is attempting to blend.
 
i'm confused. Is the difference visually noticable when you view the regular image and the "30 megapixel" image side by side at the same size?
 
I "Free Transform" the two duplicate layers by "1" Pixel each

1 pixel shift, not .5 pixel shift. Besides, even if you were to trick PS into interpolating images this way, it would be no different than simply increasing the resolution (which simply takes a pixel of a given color and turns it into x pixels).
 
i'm confused. Is the difference visually noticable when you view the regular image and the "30 megapixel" image side by side at the same size?

This is interesting. What I'm viewing this as is something like an image like a projection TV does with three color guns.

Or it could be viewed as you are making a three color separation of the images, which in effect, makes each pixel larger.

I'm going to watch this one. Interesting technique.

For the debate. It's a matter of perception, not physical changing the image into 30 mp.
 
1 pixel shift, not .5 pixel shift.

I did some tests with PS Elements 5 (my PS is only 7.0), with a 2x2 test "image" (checkerboard), and it seemed to let me position the layer with sub-pixel positioning (after jumping through a few hoops), and gave me the results that SEEMED to indicate that it was interpolating sub-pixel. However, I want to do quite a few more tests before I can be sure.

I tried it with a 10 MP image, and although it seemed to be doing *something*, I could not reasonably equate that something to "looking more like a 30 MP image". At 10MP resolutions, it all looks like a jumble of pixels close up, and it all looks fine zoomed out.

Besides, even if you were to trick PS into interpolating images this way, it would be no different than simply increasing the resolution (which simply takes a pixel of a given color and turns it into x pixels).

True, but it would give you a bit of control on HOW it's interpolated.

Unless and until Ben tells us step by step how to get the "effect", I can't really comment on it's effectiveness.

In all honesty, I probably couldn't tell a 10MP image from a 30MP image, either on a reasonable size display screen, or at a reasonable print size (ie not the side of a building)...so I wouldn't be using the effect much anyways.
 
Well essentially, even if you interpolate, you aren't actually increasing the resolution of the original image. And I think that's nuff said.
 
Sounds like we need Benji to sign back on and elaborate with everyday scenarios, explanations of what he is trying to achieve, and then before and after examples.
 
I just checked in CS2...not possible to do anything in .5 pixel increments. All pixel values have to be whole number integers.

And I tried the 2x2 thing as well, at 1600% zoom.
 
I just checked in CS2...not possible to do anything in .5 pixel increments. All pixel values have to be whole number integers.

And I tried the 2x2 thing as well, at 1600% zoom.

I can go into detail if you want, but it's just an academic exercise until Ben provides details. I say we just go out and ride bikes instead. :lmao:
 
Status
Not open for further replies.

Most reactions

Back
Top