Minimal Computing Design Principles, MSI Capture Equipment Rodeo

Minimal Computing Design Principles

Building a Multispectral Imaging System for Simplicity, Sustainability, and Accessibility

Multispectral Imaging Capture Equipment Rodeo
Rochester, New York, June 27, 2025

Todd Hanneken, St. Mary’s University

Minimal Computing

Simple
Sustainable
Accessible / affordable

It will take a few minutes to explain what I mean, but in a word, “minimal” is how I would describe my design principles. In the field of Digital Humanities there is already a vibrant discussion about the advantages of minimal computing. Some remain skeptical, or dismiss it as DH for poor people. I can see that it is not right for everyone. It is also clear that there is no one minimal approach. Rather, minimal computing is a set of questions to bring to every design decision along the way. Is there a simpler way of doing this? Is this approach going to be sustainable? Is there a more affordable or accessible way of doing this? There may be a wide variety of situations in which a more complex, short-term, or expensive option might be better. I believe it is often the case that these three principles converge, and the simplest solution is the most sustainable and accessible. Let me say a little bit more about each of these three principles before I turn to specific examples from building a multispectral imaging system.

Simple

Alignment of complexity of tool and understanding of operator
Simple tools are reusable
Self-reliance
Build it yourself
Customize to your own needs

The first principle might be the most contentious by itself. I can easily imagine that a sales representative from PhaseOne would insist that their design principles are simpler than mine. Here's a pretty machine that lets you push a pretty button and take a pretty picture. It's so easy a monkey could do it. You don't need to understand anything about what you are doing. My approach to simplicity is not that the machine needs to be fool proof, or that the target user needs to be a fool. My approach is to say that there should be alignment of the tool and the operator. If the operator has a basic understanding, the operator should start with basic tools. If the tool must be more complex, then the operator must develop a more complex understanding. I learned how to use a screwdriver and a hammer at a very young age. I have learned how to use more complex tools over the years, but I have not abandoned need of a hammer and screwdriver. I use a screwdriver for many things besides driving screws. Simple tools are the most reusable. I'm not just talking about physical tools, but skills. When there is alignment between the complexity of the tool and the understanding of the operator, we achieve self-reliance. The other principles of sustainability and accessibility follow naturally. I want to clarify here that when I talk about self-reliance I do not mean individualism at the expense of collaboration. Long-term collaborators that contribute different skills and will not go away when grant funding gets cancelled are part of a collective self-reliance.

Sustainable

Adaptable
Serviceable
Experimentation

Accessible

Affordable
The best camera is the camera you have with you
Available

I deliberately put affordable last because I resent the perception that Minimal Computing is just “Digital Humanities for poor people.” It's not false, though. I would point out that many of the institutions with the largest budgets will still complain about how they are under-resourced. If minimal computing is DH for under-resourced people, maybe that means it is for everyone. Note that “affordable” is not an absolute number. In my current stage of life and academic position I can come up with a few thousand dollars for a camera that I would not have considered possible twenty years ago. Working on building a system over the past year, I have often thought about the saying, “the best camera is the camera you have,” or “the camera you have with you.” It may be partly about portability, but it is largely about accessibility more generally and doing what you can with what you have or can afford. I think it is great that all my students have cameras with them all the time built into their phones. It is only a couple of bucks to buy a UV flashlight for fluorescence. A mount might be worth a few bucks more. Learning how to capture raw and process registered images is worth a few hours of time. Not having thousands of dollars to spend is not an excuse to not get started. Another principle of accessibility is that accessible, affordable, and available often correlate. Specialized parts are expensive and hard to find. Mass-market parts are cheap and easy to find. The system I demonstrated this morning included parts from eBay and parts I already had. When we use standard parts we will have an easier time of finding replacements when needed.

Weaknesses

Iterative learning process takes time
So-called learning curve

Before moving on to specific examples of questions I faced and choices I made when building a system according to minimal computing principles, I would like to acknowledge a limitation or two. First, the iterative learning process I am assuming takes time. It is true that if you only have one shot at doing something right, and if you can get a big grant to pay for it, the systems or services that are designed to be fool-proof might be your best bet. You have a better chance of success on the first or only try. If you are not successful, you can at least say you tried the best possible, and therefore success is impossible. There can be a lingering uncertainty when we assume that a method less than the very best is usually good enough. I am reminded of IBM’s controversial marketing campaign, “No one ever got fired for buying IBM.” It was effective because it appealed to an instinct for individual self-preservation by decision makers. At the same time, it seemed to concede that that was the only reason to choose IBM, while a little risk tolerance would lead to less expensive, equally sufficient, even more innovative options. I would be a hypocrite if I did not acknowledge that I made the same basic choice for the one palimpsest most important for my own work. For other projects, the not-quite-once-in-a-lifetime imaging projects, I believe simpler options are worth considering. Another critique of the minimal-computing approach, a critique which I only partly accept, is the argument from the “the learning curve.” I find this critique is often incredibly subjective. I have heard this phrase many times when it was far from clear that the proposed approach was any easier to learn for anyone besides the inventor. My bigger problem with the critique is that it seems to assume that learning is a bad thing. I think it is a real problem when people think that earning a PhD and an academic position means we are done learning, and now only tell others to do so. I have had to advise theology majors struggling to pass the mathematics requirement for their liberal arts degree. The learning curve was not gentle for them, but they had to do it. My attitude toward learning is to not seek ways to avoid it, but to distinguish what is most worthwhile to learn. Skills that are reusable are worth learning.

Tripods

Cheaper than Kaiser copy stands
Already had two, bought a third with ability to shoot downward for less than $100
Can upgrade to more stable tripods in the future without redoing the whole design
Can shoot and illuminate horizontally
Can borrow or buy them in any city

Having described the general principles, let me describe some of the specific choices I made when building a multispectral imaging system. Of course others might follow the same principles and come to different conclusions. Adaptability, the point that one size does not fit all, is part of the point of minimal computing. So starting with the support, I went with mass-market tripods for affordability, adaptability, and availability. They are mass-market and easily available to borrow or purchase new or used at a variety of price points. The basic principles are universal, so upgrading a component later will not impact the rest of the design. I place a high value on being able to image targets of different sizes and orientations, including wall art.

Sensors

Google Pixel 6 (FIYHI=free if you have it), added a $5 tripod mount
Canon T1i (FIYHI), no modifications for infrared
QHY mini ($509 including filter wheel), panchromatic, cooled, 8 megapixels
QHY 600 ($3909 + $255 filter wheel), 61 megapixels

With sensors, there was not a single obvious answer. I still think a cell phone is a good option in many situations. The Canon T1i has some advantages in the same basic category of free and visible light only. For a true panchromatic sensor with thermoelectric cooling, I think the QHY mini is a great bargain at $500. Astrophotography is not as large a market as general photography, but it is larger than cultural heritage. The main limitation is the 8 megapixel spatial resolution. That is still more than can fit on my best 4k screen. For testing and teaching, it is indistinct from the more expensive astrophotography cameras. To be able to shoot a full manuscript page at 600 ppi, I went for the 61 megapixel QHY 600. I realize $4000 is not going to be affordable to everyone, but not everyone needs to own one. Collaboration is a great way to get projects done.

Lenses

Canon APS-C (FIYHI), vignettes on full-frame sensor, significant chromatic aberration in infrared
Zeiss Milvus ($454 used), mass market, excellent in visible, very good in infrared
Macro focus rail ($80), focus stacking option
Coastal Optics (free with good friends), UV-VIS-IR apochromatic

I also learned quite a bit about lenses, MTF, and lens geometry. First, I wanted to see if the lenses I already had could suffice. I even had hope that a cheap lens might pass more UV than expensive lenses with lots of glass and coatings. Overall that does not seem to be the case. For $454 on the used market, I was able to get a much better lens. The Zeiss Milvus is a mass-market general photography lens with excellent avoidance of chromatic aberration in the visible spectrum. Infrared was not part of the marketing rhetoric, but our tests confirmed that excellent in the visible also means very good in the infrared. I do not picture myself buying a true specialized UV-VIS-IR apochromatic lens anytime soon, but I mention the Coastal Optics to illustrate the value of friends and collaboration.

Filters

Kodak Wratten 2 filters are reasonably mass-market, available on eBay for $20-40 each
Blue, Green, and Red for color and UV-induced fluorescence
Two infrared pass filters for infrared fluorescence
Lens-front visible-pass filter for outdoor color photography

Transmissive Illumination

TV or monitor with LCD panel removed (FIYHI)
Broadband white
About 8mm just the panel, more with the TV housing

Octopus Lights

Modular: each channel can be connected to any light, any position
Suitable for raking
Easily adapted
Accurate color from 6500 Kelvin, 2800 Kelvin, 405, and 475 nm combined with three filters
385 nm and 405 nm ultraviolet; 730 nm, 850 nm, and 940 nm infrared
Mass-market parts: Arduino or Raspberry Pi Pico, barrel jack cables, 60 watt 12 volt power supplies
About $400 for two

In some ways the Octopus lights are not the banner example of simplicity, but they do fall within my definition of minimal computing because I was able to design and build them myself. The part that required the most new learning for me was designing the printed circuit board. I had never done that before and am not sure I will do it again, but I did enjoy learning what I learned. Otherwise, the Octopus illustrates a number of minimal computing design principles. First, the lights are extremely modular. It is trivial to change a wavelength or move a light position for raking. It is called the Octopus because it has one controller head and eight feet that can go in eight different locations. For diffuse illumination, it is not necessarily helpful to run eight separate cables to the same array of lights. Here the advantage is that it is trivial to replace any light, simply by unplugging and plugging in a different barrel jack. Another example of a simplified approach was to reduce the number of lights within the visible range. When we capture in the visible, we are trying to do two things. First, we want accurate color. Second, we want to distinguish metamerism. The combination of three Wratten filters with a D65 illuminant already has some advantages over an array of narrowband illumination. Adding a second white at 2800 Kelvin gives us six combinations of visible reflectance data. We can get up to ten by adding 475 and 405 with the appropriate filters. I also added 385nm, 730nm, 850nm, and 940nm LEDs for non-visible illumination. One of those feet can also be dedicated to a raking shot from a different light position. The two main D65 LEDs can also be controlled individually, for a total of four raking shots. Each LED has optics that can be switched from narrow to medium focus. The idea here is to allow the lights to be positioned further away, which improves collimation and reduces color shadow. While the printed circuit board is custom, many parts follow the principles of broad availability. The controllers are Arduino based (with an option to use the Bluetooth Raspberry Pi Pico). The power supplies and barrel jack cables are very common parts. It is very easy to adapt the whole configuration.

Networked Capture Controller

Based on Raspberry Pi 5 ($100 with power supply, storage, case)
Controlling the lights, camera, and filters is the work of a device on the network
Compatible with Arduino (Octopus, Misha) and Raspberry Pi Pico for lights
Compatible with Canon, QHY, Flir, Pixelink, CHDK, and presumably other cameras and filter wheels
Written in Python with Flask for web interface

I also bought a Raspberry Pi to dedicate the work of controlling lights, camera, and filters to a device on the network. Following the principle of modularity, the device does one thing and does it well. The user is not tied to the Windows operating system or a particular location. One might connect from a phone for focus, or a MacBook around the corner to reduce screen light, or a different country. Data can be saved directly to the network for immediate analysis. It exemplifies adaptability in that I built it originally for my own cameras and lights, and adapted it to other lights and cameras in the rodeo. The cost in materials is quite low. The cost in time was non-trivial if starting from nothing, but I enjoyed learning through the process. The user interface is not going to steal customers from PhaseOne anytime soon. It is in a class by itself for the power to control all aspects of the capture. The rodeo would not have been possible without it.

Conclusion

Simple enough to understand, learn if necessary
Sustainable enough to grow, fix, experiment
Affordable in materials, value of time
- $5000 parts
- -$15000 valuable education on knowledge and skills I want for the future
- $2000 wasted education on stupid evil never want to see again Canon EOS Digital Software Development Kit
- -$8000 total cost

While some of the details of my project may be a bit unusual, I would like to bring us back to some general principles that might be useful to others. At every stage I made sure that everything was simple enough that I could understand it. At quite a few points, that required me to expand my understanding, which I happen to enjoy. Having struck a balance between simplifying my approach and expanding my understanding, I am already seeing the benefits of sustainability. My efforts, though significant, are generalizable skills that will come in handy when I adapt to the next situation. Due to the miracle of collaboration, not everyone needs to have all the skills. The cost in materials is excellent. One can do real, panchromatic MSI for less than $1000, and archival quality at around $5000. The discussion about affordability is about the value of the time it takes to learn. As a professional educator, my livelihood is based on the premise that people should pay for the privilege of learning. If learning is a value rather than a cost, I would estimate the overall cost of my system as about negative $8,000.