Create your own lens model through laser ray tracing

This is a slight edit of the original post.

In this post I will present an outline how anyone can create a 'thick lens model' of any lens.

Of course, the first question to answer is: why bother?

Do I really need a lens model to be a better photographer?

For instance, as a single shot macro photographer, you don't really need to know anything about the lens you're using, ie all you need to do is focus on the subject and capture an image, knowing that closing down the aperture will give you more 'depth of field' coverage, but at the cost of introducing diffraction blur.

If undertaking macro stacking, all you really need to know is the focal length and the magnifications. From this you can use wave optics to estimate the depth of field to inform the step size, ie the total visible band DoF being  (0.00055*4*(F*(1+M/p))^2)/M^2. With p and M being easily estimated by simple measurements.

If you are a 'rotating' pano shooter, who wishes to minimise post processing stitching issues, you will want to position your lens entrance pupil over the point of rotation. But you can do this easily by adjusting the camera-lens position on a nodal rail, by looking at the misalignment between a close and far object.

As a landscape photographer you may wish to know the hyperfocal distance, and this can be estimated from (F*F)/(N*c), where F is the focal length, N the aperture and c the circle of confusion value you have decided to adopt. But, of course, any sensible landscape photographer will focus beyond the hyperfocal. That is magnification can be ignored, and the position of the entrance pupil (from where the hyperfocal is measured) is a near irrelevance in the field, eg using the sensor position (marked on the camera body) is good enough. Some, like me, will simply use the 'Rule of Ten' to estimate the hyperfocal in their head (see the link on the right), and set focus accordingly. 

Therefore, the answer is: no, you don't really need a lens model.

But for those that still wish to have a lens model that is 'better' than using a thin lens model, ie assuming all the lens cardinal points sit a focal length from the sensor, the following shows you how to create a thick lens model at home, with equipment you likely already have.

Before looking at an outline of the process, the simplest and by far the best approach is to go to Bill Claff's excellent site (PhotonsToPhotos) and look to see if your lens is registered on his Optical Bench Hub. As a minimum, if your lens is in his data base, it will give you an ‘accurate’ (infinity focused) lens model at the focal length extremes for a zoom lens, or sometimes at intermediate zooms, (or at a single focal length for prime lens); and for some prime lenses you may also be lucky that Bill has been able to construct a focus model as well.

But if your lens is not in the PTP database, or you have a multi lens set up, ie a lens and, say, an optical extender or reducer coupled together, then you will be out of luck. In this case, the following procedure may be of interest to you.

One caveat to note is that this procedure works 'best' for 'rectilinear lenses, eg not fish eye lenses. As usual, I’m using the following thick lens model:

Establishing the Focal Length 

As a worst case, we will assume you don't know the lens infinity focal length. In this case, one simple way to estimate this is to use a macro tube extender, eg:

• Measure the extension tube, attach the extension tube between the camera body and the lens. Ensure the lens focus ring is turned all the way to the infinity symbol.
• Place a ruler in front of the lens. Move the camera (or the ruler) back and forth until the markings on the ruler are perfectly in focus.
• Once in focus, look through the viewfinder or use "Live View" to see how many millimeters of the physical ruler fill the width of your image. Or capture an image and evaluate in post.
• Estimate the focal length from (L*x)/S, where L is the extension length in (say) mm, x the length of the ruler visible in the image, and S the size of the sensor, eg 36mm for a full frame. 

Alternatively, accept the manufacturers infinity value. But note, at focus distances less than infinity, the focal length can change, especially for a macro lens.  

Using laser line ray tracing

All we need to do, to construct a first order thick lens model, is to locate the entrance pupil on the optical axis (we can not use the laser line technique to locate the exit pupil on axis location, this needs to be calculated), measure the pupil magnification, measure the focal length, and locate the sensor location from knowing the flange position of the lens and looking up the manufacturer’s published flange focal distance.

There are several ways to locate the entrance pupil, eg using the pano rotation approach or with a laser leveller as in this post . As for the pupil magnification, we simply estimate this by, say, taking an image of the exit pupil diameter and divide this by the entrance pupil diameter.

Assuming we are assessing the lens at infinity, first, set the lens focus to infinity, and, for a zoom, the focal length you wish to model. Then measure the focal kength.

My advice is to set up a camera over the scene and take images of all tne laser lines, for assessment in, say, photoshop. I use an iPhone and mark all the rays on a composite image in photoshop. It is also essential that you place a ruler next to lens, for scaling, and ideally positioned at the optical axis to remove parallax errors.

The first step is to establish the optical axis of the lens. This can be readily achieved by first drawing a line in the centre of a sheet of paper and aligning the laser leveller (using a single vertical line or cross lines) along the pencil line and tweaking the lens position and angle until the projection of the laser (the spot) from the lens aligns with the laser line that does not go through the lens. As can be seen, I make use of a simple white surface to help with this step:

The important thing now is to NOT touch the lens.

We now use the laser leveller at an angle to the optical axis, and shine it through the front of the lens, and adjust it so that the projected spot is symmetrical. This laser line is a reasonable estimate of the on axis entrance pupil location.

We now need to mark the position of the rear flange surface. I find the best way is to use two laser levellers to locate the flange surface. Knowing this allows us to add in the flange focal distance, to locate the sensor location. 

Knowing the focal length and using laser line, physical ray tracing, we have now captured all the information we need, ie the entrance pupil location on the optical axis, and the location of the sensor. Plus we know the pupil magnification and the focal length.

Putting it all together

The final step is to make use of Bill Claff's Thick Lens Optical Bench (TLOB):

However, before we can use the TLOB, we need to work out the internodal distance (I).

All we need to do then is enter the following information into the TLOB:

• Focal length (accepted or measured)
• Pupil Magnification (calculated from taking a picture of the pupils)
• Internodal distance (calculated from the measurements we took)
• N the aperture of interest 
• Focus distance or magnification of interest if mapping a lens not at infinity

The PhotonToPhotos Thick Lens Optical Bench will then give us a model of our lens. Will it be perfect? No, as the measurement accuracy we can achieve will limit things. Will it be a better model than using a thin lens model? YES!

Will it be fun and interesting to do? Once again YES!

As this has been a rather long post, I'll bring it to a conclusion, and follow it up with posts that illustrate the technique in action.

As usual I welcome any comments on this post or any of my posts. 

Geting to know your lens: The Ultra Digital XPan

In the last post I covered my 'final' experiment in creating a Digital Multi-Format Shift Camera set up, that covers XPan like sensor capture (over 65x24mm sensor equivalence)  and Medium Format sensor capture (for example a 78x44mm sensor equivalence).

In this post I will show how one can get an optical model of this system, simply by knowing a few, easily obtained, dimensions; eg the entrance pupil location on axis, the camera body's flange focal distance, the pupil magnification, and the focal length.

In this post I will look at my Canon R, coupled to the Laowa EF to RF Magic Shift Converter, coupled to the Kipon 645 to EF shift adapter, coupled to my 45mm Mamiya 645 lens, set at infinity focus.

The location of the on axis entrance pupil location of this three sub-system lens system can be estimated using my laser leveller method (see this post here). 

The set up is simplicity itself and estimating the position of the on axis entrance pupil takes a minute or so. The set up is to place the lens or lens system in this case, horizontally on a sheet of paper on a flat surface.The first step is to set up a suitable rear surface to help us adjust the laser angle, ie the cardboard box in this image. Then we use the laser leveller to establish and mark a zero from which we will measure the position of the entrance pupil location, in this case the front edge of the lens, which we will use as our reference plane. We could have also used the lens rear flange:

Having established a reference, we now simply use the laser leveller to record two rays that will allow us to locate the entrance pupil. It is good practice to not move the laser too far from the optical axis, due to potential pupil curvature:

As can be seen, one does not need to worry about focus on the rear surface, as all we are looking for is for the laser to be centred in the projection. With these three lines, we now have an estimate of the location of the on axis entrance pupil, in this case 36mm, from the front edge of the lens body:

Next we measure from the front reference edge of the Mamiya lens to the flange surface that will abut with the camera, ie the one for the rear shift Magic Shift Converter (MSC). In this case I estimated it at 138mm, and I called the overall lens diameter 75mm. That is all the lens complexity is represented by a black box model, 138mm long, 75mm in diameter, with an entrance pupil at 36mm from the front of the black box.

The next task is to estimate the pupil magnification of the lens system, which is easily done by inspecting the front and rear of the lens system. In this case, my eyeball approach told me the pupil magnification was near unity, ie the lens system was neither retro focus of telescopic in nature. That is, the MSC’s influence on the Mamiya 645 45mm, is to convert the pupil magnification from 1.6, ie a retro focus lens, to unity.

Also, although I'm using a 45mm Mamiya 645 lens, because of the MSC this becomes a 45x1.4 lens, ie a 63mm focal length lens system. From the lens depth of field scale, and the manufacturer's data, the minimum focus distance is 450mm (which I took as read).

The Canon RF system's flange focal distance is 20m, which allows us to estimate the position of the entrance pupil, relative to the sensor plane, ie 122mm.

As for the sensor, I simply used the 35mm diagonal, ie around 43.3mm, but I could also enter any of the stitched sensor dimensions, ie to estimate the angle of the field of view.

The final stage was to input all the measurements into the Poor Man's Optical Bench (see here) and focus at 'infinity', ie 60m away. Giving the following output:

 

The above providing us with some useful information, eg an estimate of the location of the no parallax point, ie the on axis entrance pupil location and from where the hyperfocal distance is referenced eg as shown below:

Bottom line: although we can tap into the PhotonsToPhotos Optical Bench, thanks to Bill Claff’s great work, in some cases our camera/lens set up can't be realised through PTPOB. In such cases, this post has shown how one can get an estimate of the main camera/lens characteristics. Surely a better model than using a thin lens model. 

As usual I welcome any comments on this post or any of my posts.

 

 

 

 

 

Ultra 'Digital XPan' Image Quality and Multi-Format thoughts

In the last two posts I discussed my Ultra Digital XPan set up: using a Canon RF camera, a visible band R and an IR converted RP in my case, combined with Mamiya 645 film primes (35mm, 45mm, 80mm and 150mm), coupled together via two shift adapters - one of them with optics that result in a stop increase in focal length and aperture.

As photographers we are conditioned mainly through our available technology. For example the lens focal lengths and the sensor aspect ratio and size. which lead us to seeing the world through the system's 'field of view'.

For full frame cameras we 'see' through a 3:2 aspect ratio (AR) sensor. Whereas micro four thirds and digital medium format shooters usually see through a 4:3 AR.

As I evolved my Ultra Digital XPan I also saw it as an adaptable AR system. That is, not restricted to a 65:24 XPan AR.

Let's take the standard full frame sensor of 36x24mm as our baseline 3:2 AR and compare this to one of the best medium format camera systems, ie the Phase One IQ family. The Phase One sensor size being 53.4x40mm:

If we now add in the Ultra Digital XPan (flat stitched) sensor equivalent sizes, after using both adapters together, we see the following comparison:

One of the 'advantages' the Ultra Digital XPan set up is that we can rotate the camera (landscape/horizontal or portrait/vertical orientation) and independently rotate each adapter, giving the system multiple 'degrees of freedom'. We can also replace the Magic Shift Converter's optics with a 'simple' pass through adapter. Or, for example, we could 'just' use the MSC on its own, giving a 56x24mm or a 44x36mm stitched sensor equivalence.

As a final illustration of the sensor size adaptability of the set up, if we use the front shift in landscape mode of 30mm, which gets enlarged by the MSC by 1.4, and use the MSC's shift of 20mm in portrait mode, we arrive at a 78x44mm sensor equivalence, which, when compared to 35mm full frame and the medium format Phase One, looks like this:

To illustrate this 78x44 option, here is a (scaled for display) test image, taken with my 45mm Mamiya at 45x1.4mm because of the MSC, giving a 14903x8124 stitched image in Lightroom, ie some 135MP image:

Of course all the above adaptability must come at some cost.

I will ignore the obvious cost of buying the two adapters and a set of medium format lenses. A more important çost for many photographers will be the impact on Image Quality, but this is where it gets a bit subjective.

If you are a pin hole film shooter, your definition of IQ will be different to someone shooting with a Phase One IQ4 system. Thus I lean towards IQ being in the eye of the beholder. If it works for you, then all is right with the world :-)

After a very helpful exchange on the S&T DPReview forum (many thanks to those that chimed in) I decided to approach IQ in a very pragmatic way.

First, I decided to not 'worry' about resolution, as I can't change this, ie I'm stuck with using the Mamiya 645 film primes and the optics in the Magic Shift Converter. This decision allowed me to not get diverted by creating MTFs and using MTFMapper etc.

But I still wanted a 'feel' for how 'good' my set up was: so I elected to use a 'field map' approach:

• Set the aperture fully open
• Point the camera down at a flat surface with texture, eg a carpet (see set up image below)
• Place a reference object in the middle of the frame, to aid stitching. I used a ruler
• Set exposure and focus
• Capture your images for assessment
• Stitch if evaluating a pano frame
• Use the find edges filter in Photoshop to visualise the ‘IQ' across the image

The set up looked like this:

The RAW capture from one of centre pano captures looked like this:

After stitching, in this case using the full shift of both adapters with my 45mm Mamiya 645 lens at f/8, the resultant, Lightroom, (perspective) stitched capture (21216x4741) looks like this:
 

After applying the find edges filter I got this:

The stitched image in portrait mode, ie 86x36mm, looks like this:

On inspection, I am pretty content with what I'm seeing in both images. Of course there is some 'fall off' towards the edges, especially in the extreme case. But the overall linearity and ‘IQ', to me, looks OK.

As usual I would welcome any comments on this post, or any of my posts.

Ultra Digital XPan - Part 2

In the last post I introduced my 'Ultra Digital XPan' set up, which looks like this in portrait mode:

 

Using my geared head, on a good tripod, gives me solid positional control over composition. To complement the internal levelling guide in my Canon R, I've added a three axis cold shoe bubble level. Finally, I use my H&Y Revoring filter holder, to handle my 82mm filters, thus allowing me to add front filters, for example, on my 720nm IR converted RP I can add a 830nm cut filter.

As mentioned in the previous post, using dual shifts, in conjunction with the Magic Shift Converter optics, I can realise shifts of some 62mm, ie over 30mm in each direction.This allows me, as shown in the last post, to create 4:1 near parallax free captures. I say near, as the lens does move relative to the camera on one of the adapters. 

The resultant stitched image being equivalent to a single image being taken with a sensor that is 98mm x 24mm.

In this post I will discuss using the set up, in the field, in 'true' XPan aspect ratio mode; that is seeking to maximise the pixel capture. 

To achieve this, as shown in the picture above, the camera is placed in portrait capture orientation, to ensure a 36mm sensor height, compared to an XPan negative of some 24mm.

As I was feeling lazy, the test subject is a view 50m from my house. I was using my Canon R and my 80mm Mamiya 645 lens.

Having explored the exposure extremes with my camera's histogram, ie shifting across the full shift range, setting focus and aperature (f/11 in this case), I took a seven image pano bracket set that looks like this:

The two extreme captures represent the 10mm shifting using the Magic Shift Converter, whilst the other captures cover the physical 30mm shift of the Kipon 645 to EF shift adapter, with the MSC at zero. The Kipon shift then ends up as 30*1.4 mm after passing through the MSC adapter’s optics.

After preprocessing in Lightroom, using DXO  PureRaw 5 in this case, I processed the pano bracket set using Lightroom’s Photo Merge, using Perspective projection.

This resulted in a 16165x6782 image, equivalent to a 86x36mm sensor, ie larger than a classic film XPan negative, of 65x24mm.

After cropping the image to an XPan format (65:24), the final, digitally mounted, image looks like this (greatly scaled here because of its size):

To give a feel for the details, here is a 1:1 screen capture of a part of the image:

As far as I'm concerned the experiment has been a great success and I'm content I've got a usable digital XPan-like set up, albeit using medium format Mamiya film lenses and sensor bracketing.

As usual I would welcome any comments on the above or any of my posts. 

 

Ultra Digital XPan - Part 1

If you have been following this blog, you will know I like to ‘play about' with adapters; especially those that allow me to capture images for perfect flat stitching in post.

In previous posts, for example here, I have discussed the Laowa Magic Shift Converter and illustrated its potential when coupled to a Canon TSE 24mm Tilt/Shift lens.

In this post I'll discuss a more flexible arrangement, that allows different focal lengths to be used.

The lenses I've explored are from the Mamiya 645 family, ie 35mm, 45mm, 80mm and 150mm. Although I've dropped using the 35mm as, unfortunately, I seem to have a bad copy, with one side of the image rather distorted, compared to the other side.

In this test I used the 45mm Mamiya 645 lens.

The  camera set up I used loks like this:

That is: a Canon R, coupled to the Laowa EF to RF Magic Shift Converter, coupled to the Kipon 645 to EF shift adapter, coupled to my 45mm Mamiya 645 lens. With this arrangement my total shift is of the order of 50mm, ie 25mm in each direction.

Of course the MSC transforms the 45mm focal length into about a 63mm lens.

Having set up the composition, checked the exposure over the shift range, set focus and aperature (f/11 in this test), it was a simple matter to capture 5 images:

• Kipon +15mm + MSC 10mm
• Kipon + 15mm + MSC 0mm
• Kipon 0mm + MSC 0mm
• Kipon -15mm - MSC 0mm
• Kipon -15mm -  MSC 10mm

Once captured, the five images were processed and Pano mergered in Lightroom. The resultant (flat stitched) pano was 18636 x 4664, ie just short of 87 mega pixels.

This is achievable because of the optics in the MSC, which expands the Kipon’s 30mm physical maximum shift by 1.4.

At the Canon R's pixel pitch of 5.34 microns, this is like using a camera with nearly a 100mm x 24.9mm sensor, ie 36 + 30*1.4 + 20. Graphically looking like the comparison below: where the red rectangle represents a normal full frame sensor, and the blue the 'Ultra Digital Xpan' created with the above set up.

As for the test image, it looks like this (scaled to display in the blog). The 35mm full frame, horizontal, field of view equivalent is similar to a 23mm focal length:

So, I think, finally, I've found my ideal digital XPan set up. If I want to create an ultra wide, 4:1, flat stitched, pano I can use the MSC with the Kipon shift adapter, converting my Mamiya lenses into a 49mm, 63mm, 112mm, 210mm set . If I only want to emulate an XPan aspect ratio, I would replace the MSC with a simple EOS to R adapter (which I have), allowing me to shoot at 45mm, 80mm and 150mm (as my 35mm is not really useable). Finally, with the MSC arrangement I could also shift in two directions using the Kipon and MSC in combination, ie to create various medium format equivalent aspect ratios.

As usual I welcome any comments on this post or any of my posts.

 

Digital XPan field report

In previous posts I’ve covered various ways to sensor bracket, and how to create a digital XPan DNG, by flat stitching a pano equivalent of a 65x24mm film negative. 

In this post I’m simply going to show a few test shots that I just took with my Mamiya 645 45mm medium format film lens, attached to my Canon IR converted RP, via one of my Fotodiox, +/- 15mm, adapters (use the link on the right to explore the various adapters that Fotodiox have).

As usual I first used one of my iPhone Apps to help me with framing, in this case the ViewFinder Mk 2 iPhone App: https://www.artistsviewfinder.com/

In the above we see three of the lenses I have.

Another App I use is the Viewfinder Preview: https://viewfinderpreview.opticalaberration.com/

Post processing was carried out in Lightroom and Photoshop. 

In this particular case, I decided to process the images in black and white. Two were processed at an XPan size, ie shifting in landscape mode, and one in a medium format 5x7 size, ie shifting by +/- 15mm in portrait mode.

To show off the images on line, I digitally mounted them using the DBW plugin, that you can purchase from here. 

I hope this short post has piqued your interest in sensor bracketing and camera format shifting. If you can’t afford an XPan or don’t wish to get into film photography, using a second hand medium format lens and a format adapter, is a great digital alternative.

As usual, I welcome any comments on this post or any of my posts. 

 

Double Adaption: for ultimate control in creating a ‘digital XPan’

Those that have been following my recent posts will know that I've been experimenting with sensor bracketing, and, in particular, the 'best' way to simulate a digital XPan, using one of my Canon RF or EF-M cameras (either visible or IR) and various adapters, eg see the Fotodiox site for examples of the many adapters you can get.

In this post I'll be looking at the benifits of double adaption. That is using more that one adapter.

As discussed previously, using Mamiya 645 lenses, with a larger image circle than the Canon RF format, allows one to exploit shift, and one might be tempted to opt for a single adapter, such as the Fotodiox Mamiya 645 (M645) to Canon RF Shift Adapter:

This adapter allows some 30mm of shifting (+/- 15mm); which initially looks great, until you factor in the Mamiya lenses.

As an example, lets look at the 35mm Mamiya - Sekor C f/3.5 N that I have (#107008):

Like all/most Mamiya 645 lenses, the infinity is at a hard, mechanical, stop, and my experience is that infinity can be slightly thrown out when coupled with adapters. In theory one could try and adjust the lens for a more accurate infinity with an adapter, however, this could end up as a wasted effort if you are using different adapters, as each adapter may be a few microns adrift from each other because of manufacturer’s design, quality control and tolerances.

This is where double adaption comes in.

The first adaption is to introduce a shift adapter to go from a Mamiya 645 lens to a Canon EF mount, then add a focus correction adapter to go from EF to RF. The adapters I have used are the Kipon Shift M645 - EOS, and the Fotodiox dlx Stretch EF - EOS R.

 

As can be seen, the additional advantage of using the EF - RF stretch is that it comes with rear, drop in, magnetic ND filters (ND4, 8 and 16). The final double adapter set up looking like this:

Using this arrangement of adapters allows me to capture a 61.75 x 24 mm, stitched, flat, panorama, using my visible band R or my IR (720nm) converted RP. That is creating a digital XPan image, albeit by using a medium format Mamiya film lens, rather than an XPan film lens. 

The in field workflow I've come up with goes like this:

• Set the Mamiya on its widest aperture, eg f/3.5 for my 35mm lens;
• Calibrate the lens for infinity by putting the lens at infinity, ie zooming in to the farthest point of interest in the image, and adjust for maximum focus at that point, using the stretch adapter;
• Set the required aperture (I tend to use f/16 or f/22 for maximum depth of field, and correct for diffraction softening in post) on the lens and the exposure time, using the histogram and ETTRing;
• Ensure the ETTR exposure covers the shift extremes;
• To ensure maximun pano blending success, I personally take three images, ie at the two extremes (+/- 15mm) and one in the middle with zero shift.

As for post processing I find the following workflow creates what I'm looking for:

• Ingest into Lightroom;
• Use Raw Detail to clean up the images;
• Use pano photo merge;
• Use Topaz AI, including, if required, dust and stratch (although this is now available in Photoshop ACR and hopefully will be in Lightroom soon) and Topaz Superfocus, to tidy up any diffraction softening;
• Tone and finish as required.

The following test shot was taken with the 35mm Mamiya at f/22, and I've zoomed in to show the detail that I obtained. The Image was taken with my IR converted RP and the final image is 10772 x 4217, ie 61.7mm x 24.2mm, and left in RGB space:

The bottom image above is a screen grab from a 1:1 zoom, to illustrate the details.

For completeness here is the B&W version, which I prefer, as colour integrity can be a challenge when post processing IR captures in colour:

I'll wrap up this post at this point as I think I've covered everything I wished to discuss. The bottom line being that double adaption is a viable way to ensure infinity focus on a manual MF lens when sensor bracketing. For me, knowing I can simulate an XPan with my mirrorless digital camera is fun and, I hope, will turn out to be worth while.

As usual I welcome any comments on this post or any of my posts.