What to do when a 10 billion Pixel scan is too small?

[DavyC]

What to do when a 10 billion Pixel scan is too small?
Easy, redo it at 108 billion!

So, it was with the Johannes Vermeer painting, ‘Girl with the pearl earring’.

https://www.youtube.com/watch?v=j_MvpMlgfwI

[bobfriedman]

very interesting... thx for sharing.

[rjlittlefield]

Interesting and lovely!

That technique of "elevated printing", to produce a 3D enlargement of the print surface, is new to me. (Sequence starting at 4:30 in the video.)

More information about elevated printing in this brochure by Canon: "PRISMAelevate XL".

--Rik

[DavyC]

[quote
very interesting... thx for sharing.
[/quote]

Yes Bob, 88 hours for a scan!
That's dedication.
I had some 'Gear Envy' watching this.
D.

[DavyC]

More information about elevated printing in this brochure by Canon: "PRISMAelevate XL".

Cool process. Going to be people putting their fingers all over the work for some tactile feedback though.
D.

[canonian]

Ah, great to see my good friend Emilien Leonhardt, with whom I have spend plenty of fun hours playing with the incredible Hirox 3D microscope.
Although with a 5 to 6 digit pricetag (depending on which couple of accessoires and extentions you add) I will likely never own one...

[ray_parkhurst]

I'm trying to figure out the magnification for this work. It was stated that the original smaller image was scanned at 35x, and the new one at 90x.

It was further stated that it took 41000 images (I assume this was after stacking) to complete the scan. The painting is 44.5 x 39 cm, or 1735.5 sqcm, or 173550 sqmm. This means that each image represents 173550/41000=4.2 sqmm. Assuming a square sensor for convenience, this would image a 2mmx2mm area of the painting. At 90x magnification, the sensor would need to be 180mm x 180mm, which seems unlikely.

Looking at this from another perspective, the final image is said to be 108 Gigapixels. This means each tile is 108e9 / 41e3 = 2.6MP after tiling. Hmm.

Checking on the HiRox website, indeed their HRX-01 is offered with two sensor options...a 3MP and a 5MP, both CMOS.

It would seem that the "90x" magnification is a "marketing" magnification, similar to how most USB scopes are marketed. Indeed looking at the HiRox lens product page, they offer lenses up to 10,000x, so it is a bit difficult to say exactly what magnification the "Girl" was imaged with. If we assume a 2/3" sensor (11x6.6mm, 72.6sqmm) then the actual optical magnification used would be around 4x.

[DavyC]

Ray,
Here is the link to the interactive Vermeer art scan.
Providing Ruler tools etc. Pretty impressive stuff.

https://www.hirox-europe.com/gigapixel/ ... KPROCESSED

[ray_parkhurst]

It really is a very impressive work.

Thanks for the link to the viewer @DavyC. The Hirox viewer is similar to the EasyZoom viewer, though the Hirox updates more smoothly and has a few more features.

I found recently that the 95PN I've been using at 1.7x to do stack and stitch of Lincoln Cents has shallow enough DOF to do decent 3D renderings. I shot the last few with more DOF overlap in order to smooth the 3D, and crops from them rendered well. Certainly the "Girl" shot at 4x can do the same thing, though I only do 2 tiles, not 41000. With only 2 tiles, I can crop pretty much any size or shape and do a 3D rendering. I'm wondering what the limitations are for 3D of the "Girl". For the 2D image, I'd expect them to have stacked each tile, then stitched the panorama. It is not practical on such a large scale to stitch each stack slice, then stack the stitched images. I suppose this limits them to doing 3D rendering of fairly small (~2.4x1.6 mm) tiles, though for specific areas they could stitch a small patch of tile slices and then 3D render that patch.

[DavyC]

"It really is a very impressive work"

Ray,
I’m not too clued up on 3D rendering. I need to do some reading up on that.
Was recently working with a 2D museum scan, so this got my interest.
Not so many pixels, but the scale was pretty big (35,052,000 square mm) so >200 times the area of the Vermeer painting.

Doing things, the hard way. Manual stitching, individual colour corrections, tweaking etc. It was heading for something like 350,000 pixels long.
It was becoming unwieldly and taking days. The goal in the end was just to make a simple moving HD mp4 video.
With Photoshop refusing to play with jpegs over 30,000 pixels, I ended up cutting things down to a more manageable 90,000 x 1080 pixels.

Video lasted for around 9 minutes, simple panning from left to right.
D.

[ray_parkhurst]

"It really is a very impressive work"

Ray,
I’m not too clued up on 3D rendering. I need to do some reading up on that.
Was recently working with a 2D museum scan, so this got my interest.
Not so many pixels, but the scale was pretty big (35,052,000 square mm) so >200 times the area of the Vermeer painting.

Doing things, the hard way. Manual stitching, individual colour corrections, tweaking etc. It was heading for something like 350,000 pixels long.
It was becoming unwieldly and taking days. The goal in the end was just to make a simple moving HD mp4 video.
With Photoshop refusing to play with jpegs over 30,000 pixels, I ended up cutting things down to a more manageable 90,000 x 1080 pixels.

Video lasted for around 9 minutes, simple panning from left to right.
D.

My 3D rendering is limited to commercial programs, so no expert here. My goals are quite a bit more modest than what you are describing, but I do understand the basics of the workflow and its limitations, given no ability to customize.

Keep in mind this was a huge commission for Hirox, and they have dedicated programmers who create the code for stacking/stitching/presentation. Kudos to you for completing such an ambitious project. Working with large files is very difficult from software and hardware perspectives without customization and optimization.

[Sym P. le]

3 model to choose from

PRISMAelevate XL – Alto
• Textured printing: up to 1 mm
• 36 layers maximum
• Build elevation with all colors or white only

PRISMAelevate XL – Alto-Plus
• Textured printing: up to 2 mm
• 54 layers maximum
• Layers of dark ink
• Mold mode for increased speed

PRISMAelevate XL – Brila
• Transparent Elevation: up to 0.25 mm
• 18 layers maximum
• Varnish on top of the color image layer

[emilien]

Thanks for your kind comments! Glad you like the results, it was a crazy project and we learned a lot from it! I see there are some questions, see my answer below.

I'm not trying to figure out the magnification for this work. It was stated that the original smaller image was scanned at 35x, and the new one at 90x.

Looking at this from another perspective, the final image is said to be 108 Gigapixels. This means each tile is 108e9 / 41e3 = 2.6MP after tiling. Hmm.

Checking on the HiRox website, indeed their HRX-01 is offered with two sensor options...a 3MP and a 5MP, both CMOS.

It would seem that the "90x" magnification is a "marketing" magnification, similar to how most USB scopes are marketed. Indeed looking at the HiRox lens product page, they offer lenses up to 10,000x, so it is a bit difficult to say exactly what magnification the "Girl" was imaged with. If we assume a 2/3" sensor (11x6.6mm, 72.6sqmm) then the actual optical magnification used would be around 4x.

Hi Ray,
Thanks for your question!
The way to calculate magnification is always a complex subject and I understand the confusion. First of all, Hirox is the inventor of video microscopy in 1985, the first company that developed a microscope using a camera, a lens and a screen without using binoculars / or the eyes as a reference. At the time, analogue monitors were mostly 15” and we didn’t use objectives coming from traditional optical microscopy, the reference that was developed was: how much time bigger is the object on a 15” screen. Once this was calculated, we developed a table showing magnification and field of view for our various lenses and kept it since then. Size of screens is not relevant and has no impact on the calculation anymore but the FOV / Zoom is still in use. There is a table here showing this ratio (FAQ3 and FAQ4): https://hirox-europe.com/support/faq/
Magnification is not relevant anyways, we are not selling our system because we can “zoom more”. What is important FOV and pixel size, image quality and resolution, depth of field (large for easy inspection / small for 3D depth measurements).
In this case, at 90x we have a pixel size of about 1.3 um and a horizontal FOV of approximately 3 mm for 2500x2050 pixels. Keep in mind that for the Girl with a Pearl Earring, we used a crop: we only captured the central part of the image, using 2000x2000 pixels for each tile. Also I used a 20% overlap in X and 20% overlap in Y. I will let you do the math!
In any case, I’m not sure how you end up with your calculation of 4x but please don’t compare our fully motorised XYZ high end system with a triple objective telecentric lens to a plastic USB microscope
I hope this clarifies some doubts and if there are any further questions, I’m happy to answer them!
All the best, Emilien

[ray_parkhurst]

Thanks for your question!
The way to calculate magnification is always a complex subject and I understand the confusion. First of all, Hirox is the inventor of video microscopy in 1985, the first company that developed a microscope using a camera, a lens and a screen without using binoculars / or the eyes as a reference. At the time, analogue monitors were mostly 15” and we didn’t use objectives coming from traditional optical microscopy, the reference that was developed was: how much time bigger is the object on a 15” screen. Once this was calculated, we developed a table showing magnification and field of view for our various lenses and kept it since then. Size of screens is not relevant and has no impact on the calculation anymore but the FOV / Zoom is still in use. There is a table here showing this ratio (FAQ3 and FAQ4): https://hirox-europe.com/support/faq/
Magnification is not relevant anyways, we are not selling our system because we can “zoom more”. What is important FOV and pixel size, image quality and resolution, depth of field (large for easy inspection / small for 3D depth measurements).
In this case, at 90x we have a pixel size of about 1.3 um and a horizontal FOV of approximately 3 mm for 2500x2050 pixels. Keep in mind that for the Girl with a Pearl Earring, we used a crop: we only captured the central part of the image, using 2000x2000 pixels for each tile. Also I used a 20% overlap in X and 20% overlap in Y. I will let you do the math!
In any case, I’m not sure how you end up with your calculation of 4x but please don’t compare our fully motorised XYZ high end system with a triple objective telecentric lens to a plastic USB microscope
I hope this clarifies some doubts and if there are any further questions, I’m happy to answer them!
All the best, Emilien

Hi Emilien,

Thanks for the further input, which confirms what I was saying, ie that the "magnification" is much lower than stated, similar to that used by the typical USB microscope. The microscope itself is of course much higher quality, so no comparison on the hardware, just in the method used to quote magnification.

As far as the numbers go, it seems you're using a much smaller sensor than I expected, and thus at a much lower magnification, closer to 1x.

It is important to know the optical magnification for calculating things like FOV, DOF, and diffraction effects. When you quote a magnification like 90x, it implies extremely thin DOF, and extremely small FOV. Indeed a 90x magnification using your sensor would only give a ~29um FOV, and would require ~8100 times more stacked images, each one with ~20 times more shots in each stack, or a total of ~162000 times more total shots required to capture the "Girl". As you can see, quoting 90x is indeed extremely confusing, and obviously could not be correct, hence my post.

I still would like to know the capability of your software for doing 3D rendering. Am I correct that you stack first, then tile after? Does this limit you to a 2mm FOV for your 3D renderings, or can you pull multiple tiles together to do 3D on a larger FOV?

Thanks again for the detailed reply, your excellent work, and best regards to you,

Ray

[Scarodactyl]

. First of all, Hirox is the inventor of video microscopy in 1985, the first company that developed a microscope using a camera, a lens and a screen without using binoculars / or the eyes as a reference.

Bausch and lomb demo'ed the monozoom 4, which was a video only microscope and the unreleased predecessor to the monozoom 7, at a convention in 1979. The monozoom 7 came out as a product at least as early as 1985. Not that either is a super incredible lens but the claim reminds me of Olympus claiming to have made the first zooming stereo microscope (years after Bausch and Lomb had debuted the actual first).
It might be cynical of me but it sure feels like the goal is a higher magnification number to sound better, not clear communication (I have only ever see these inflated numbers cause confusion, with experienced and inexperienced users alike). Not that I can really cast stones, I've totally given in to quoting the extra 10x 'hypothetical eyepiece' factor when quoting magnification in my videos to prevent internet randos from complaining.
I had a couple old Hirox lenses for a bit, great construction. One day I'd love to own one of the more modern ones.