Most people can tell the difference between a good quality image and a bad one.  The blocky 240p YouTube video vs. the crisp 1080p or 4K version, a decades-old Polaroid photo vs. a portrait off of a new high-end phone or DSLR camera. As Albedo is in the business of delivering the highest quality commercial imagery from space, understanding and quantifying image quality is crucial to our work.  Netflix has invested and published some excellent work in how they bridge the subjective and quantified in measuring streaming video quality (a related but different problem from still imagery).

Imagery has three quantifiable attributes that contribute to its subjective perceived quality: resolution, blurriness, and noise.

• You know noise from a phone camera image you’ve taken in low lighting conditions, the speckley noise in the dark areas.  Underexposed imagery is noisy and lower quality.  High reflectivity mirrors and high sensitivity camera sensors reduce noise.

• Blurriness you’re familiar with is probably motion blur from not holding your phone camera steady.  We have a similar problem on our spacecraft, where we must point with high stability to avoid smearing our imagery.  And in large telescopes used in space imagery, the aberrations introduced by the optics themselves and sampling of the telescope’s point spread function also introduce blurriness.

• Here, resolution is the distance that a pixel in the image spans.  Point your phone at your keyboard and you can have sub-millimeter resolution.  Point it at hills or mountains on the horizon, and now the same camera produces an image resolution of a hundred meters and when you try and zoom in, you cannot make out fine details.

Of the the three attributes above, resolution has a much larger impact on image quality than blurriness or noise. And before you say anything — yes, we know that this isn’t a perfect technical description. For example, take resolution: there’s temporal resolution, spectral resolution, and spatial resolution is it’s own beast. We’ve intentionally simplified this to be approachable — just like our explanation of Albedo’s great image quality: big camera, go space 😏

So, I bet you’re reading the above and thinking: “Wow, image quality sounds like a hard thing to try and objectively quantify!” Well, the U.S. Government had the exact same thought and chose to do something about it. Many years ago, different systems produced different image qualities, so the U.S. Government sought to standardize these subjective effects with a quantified scale.

Enter the NIIRS scale. If you don’t come from the government world, it might just be another acronym in your backlog of “I don’t know what this means but it’s far too late to ask”.  NIIRS — which is pronounced “nears” — stands for “National Image Interpretability Rating Scale”.  It is a logarithmic scale (like the Richter scale of earthquake magnitudes where 6.0, 7.0, and 8.0 earthquakes are very different in intensity!), spanning down to very low NIIRS 1 and 2 consistent with coarse imagery collected on the first U.S. satellites, and up to NIIRS 7 and 8 consistent with exquisite aerial reconnaissance.**

Imagery analysts, who are trained in this school of thought and earn the designation of being “NIIRS-rated”, can evaluate imagery and score it with a NIIRS rating.  The U.S. Government uses the NIIRS scale to define what quality of imagery it needs to satisfy various intelligence tasks.  But how do you know what the image quality of an image will be before you take it, or more important, before you start building your super-expensive satellite?

For this, we use Image Quality Equations (IQE) that allow you to calculate from engineering parameters what the approximate NIIRS rating will be. The General IQE v5 (GIQEv5) published by the NGA is the latest IQE, which has proven to be robust and accurate for a wide range of system parameters.  We compute the parameters that GIQEv5 uses, and can thus predict that Albedo’s image quality in good conditions will be NIIRS 7.0 (including some margin and conservatism), and this has been independently checked as well.

According to our calculations -

At NIIRS 5 and below, you can:

• Detect tennis and basketball courts in cities (NIIRS 2.1)
• Count houses in a suburb (NIIRS 2.9)

From NIIRS 6 - 7, which is the upper end of the best commercial satellite imagery:

• Distinguish between sedans and station wagons (NIIRS 5.9)
• Identify orchards by fruit type based on tree size and shape (e.g., apple, cherry, citrus). (NIIRS 6.1)
• Detect the presence of obstructions (e.g., weed growth, soil slumpage) in an irrigation system. (NIIRS 6.2)
• Identify large animals like giraffes, rhinos, and elephants (NIIRS 6.3)
• Identify pole-mounted electrical transformers in residential neighborhoods. (NIIRS 6.7)

From NIIRS 6.5 - 7 (Albedo’s Best NIIRS rating), you can:

• Measure vegetation encroachment from specific types of trees to high accuracy line (NIIRS 6.9)
• Detect a change in appearance of a very large natural gas pipeline (NIIRS 6.9)
• Detect presence of equipment pods on military aircraft (NIIRS 7.0)
• Identify medium farm animals by type (e.g., sheep, goats). (NIIRS 7.3)

At NIIRS 8 and above, which would be very ambitious for any capital-conscious startup to collect from space:

• Detect individual baby pigs (NIIRS 8.7)
• Identify an individual person wearing distinctive clothing (NIIRS 9.5)
• Read license plates on a car (NIIRS 11.9)
• Identify an individual person by their facial features (NIIRS 12.2)

The uninitiated hear satellite imagery, and immediately jump to identify an individual person thanks to popular movies and TV shows. Despite what might be reported on other sections of the internet, the NIIRS ratings above show us that it is impossible to see what clothing people are/are not wearing and impossible to detect when people are sunbathing — at least for Albedo.

Don’t believe us? Check it out for yourself: you can examine our simulated imagery, produced with a high-fidelity image chain simulation that incorporates all the effects to image quality our satellites and system introduce.

**These original definitions are very subjective, however, and as part of our AFWERX 22.1 Direct-to-Phase-II SBIR contract, we identified rating scale entries varied as much as -0.5 to +1.5 ΔNIIRS off of the subjective scale.

References for NIIRS ratings in header diagram

• https://apps.dtic.mil/sti/trecms/pdf/AD1158528.pdf
• https://www.esri.cl/content/dam/distributor-restricted/esri-cl/pdf/10073-ds-legion.pdf