💻 computer science

DEM Refinement and Validation on the Lunar Surface Using Shape-from-Shading with Chandrayaan-2 OHRC Imagery

This study presents a Shape-from-Shading framework using Chandrayaan-2 OHRC imagery to refine and generate sub-metre resolution lunar digital elevation models across three key sites, demonstrating measurable topographic enhancements while characterizing limitations related to illumination geometry and image coverage.

Original authors: Aaranay Aadi, Jai Gopal Singla, Nitant Dube

Published 2026-04-21

📖 4 min read☕ Coffee break read

CC BY 4.0

Original authors: Aaranay Aadi, Jai Gopal Singla, Nitant Dube

Original paper licensed under CC BY 4.0 (http://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

Imagine you are trying to create a 3D map of a mountain range, but you only have two photos taken from slightly different angles. This is how scientists usually make maps of the Moon: they use stereo vision. By comparing two images, they can calculate depth, much like how your two eyes help you judge distance.

However, there's a problem. If the terrain is very flat, or if the shadows are too deep (like in the Moon's polar regions), those two photos might miss tiny details. It's like trying to see the texture of a brick wall from far away; you see the wall, but you miss the individual bricks.

This paper presents a clever solution to "fill in the gaps" using a technique called Shape-from-Shading (SfS). Here is how it works, explained simply:

1. The "Lighting Change" Trick

Think of the Moon's surface like a piece of clay.

The Stereo Pair (The First Two Photos): These are like taking two photos of the clay with a flashlight held at a specific angle. You can see the big bumps and valleys.
The Third Photo (The Secret Weapon): The researchers took a third photo of the exact same spot, but this time, the "flashlight" (the Sun) was in a slightly different position.

When the light hits the clay from a new angle, different tiny cracks and ridges cast new shadows. Even if the 3D map from the first two photos was a bit "blurry" on the small details, this third photo reveals them because the light is hitting them differently.

2. The "Smart Painter" Algorithm

The computer program acts like a very smart painter.

It starts with the "blurry" 3D map made from the first two photos.
It then looks at the third photo and asks: "If my current map were correct, would the shadows look like this?"
If the answer is "No," the program tweaks the map. It raises a hill or digs a crater just enough so that the shadows in the new photo match the real image.

By doing this, the computer isn't just guessing; it's using the physics of light to "sculpt" new details into the map that weren't there before.

3. Where Did They Test This?

The team tested this on three different "playgrounds" on the Moon:

The Vikram Landing Site: Where India's lander touched down. They wanted to make sure the map was safe and detailed enough for future missions.
Mons Mouton (The South Pole): This is a mountain near the Moon's south pole. The Sun here is very low on the horizon, creating long, dramatic shadows. This is the perfect place for their "lighting trick" to work because the shadows are huge and easy to see.
Cyrillus Crater: They tested a "worst-case scenario" here. Sometimes, the third photo is taken at such a similar angle to the first two that the shadows don't change much. In this case, the trick didn't work well, teaching them the limits of the method.

4. The "Goldilocks" Settings

The computer needs to be told how much to trust the new shadows versus the old map.

If you trust the shadows too much, the map gets "noisy" (like static on an old TV).
If you trust the old map too much, you don't see any new details.
The researchers found the "Goldilocks" setting: A specific balance where the map gets sharper and more detailed without becoming messy. They found that for their high-resolution cameras, a specific setting worked best to reveal tiny craters and slopes that were previously invisible.

Why Does This Matter?

This is a big deal for space exploration.

Safety: Future astronauts or robots need to know if there is a tiny rock or a hidden crater that could trip them up. This method finds those hidden dangers.
Science: It helps scientists understand how the Moon's surface was formed by seeing the tiniest details.
Efficiency: Instead of waiting for a new mission to take better photos, we can use existing photos and this "lighting trick" to get better maps right now.

In a nutshell: The researchers took a high-resolution 3D map of the Moon, gave it a "second look" using a photo taken with the Sun in a different spot, and used the new shadows to carve out tiny, previously invisible details. It's like using a new angle of light to reveal the hidden texture of a sculpture.

1. Problem Statement

Accurate topographic models (Digital Elevation Models or DEMs) of the lunar surface are critical for geological mapping, landing site hazard assessment, and terrain analysis. While orbital stereophotogrammetry is the standard method for creating these models, it faces inherent limitations:

Spatial Frequency Limits: The recoverable detail is constrained by sensor ground sampling distance (GSD) and stereo baseline geometry.
Featureless/Shadowed Terrain: Image matching often fails in areas with low texture or deep shadows (common in polar regions).
Resolution Gap: Existing high-resolution datasets (like Chandrayaan-2 OHRC) can be refined further to capture sub-meter features that stereo pairs miss.

The paper addresses the need to enhance sub-meter resolution DEMs beyond the limits of stereo reconstruction by utilizing Shape-from-Shading (SfS) as a source of new topographic data rather than just a smoothing tool.

2. Methodology

The authors developed an end-to-end SfS framework integrated into the Ames Stereo Pipeline (ASP) and ISIS toolchain, specifically tailored for Chandrayaan-2 Orbiter High Resolution Camera (OHRC) imagery (0.25 m/pixel).

A. Data Strategy: The Three-Image Approach

Unlike traditional SfS which often refines a stereo pair using one of the same images, this study employs a three-image strategy:

Stereo Pair ( $I_1, I_2$ ): Used to generate an initial DEM ( $z_0$ ) via standard photogrammetry.
Independent Shading Image ( $I_3$ ): An independent OHRC acquisition of the same region with different viewing and illumination geometry.
- Significance: Because $I_3$ is independent, the topographic details recovered by SfS are new information, not just artifacts of the stereo pair.

B. Processing Pipeline

Initialization: $I_1$ and $I_2$ are processed to create $z_0$ . This is vertically aligned to a reference NAC DTM to correct biases.
Map Projection: The independent image $I_3$ is map-projected onto the initial DEM $z_0$ to ensure spatial alignment.
SfS Formulation:
- Reflectance Model: Uses the Lunar-Lambertian model to account for non-Lambertian effects on airless bodies.
- Optimization: Minimizes a variational energy function $E(z)$ $E (z)$ comprising:
  - Data Fidelity: Difference between observed brightness and rendered brightness.
  - Smoothness Constraint: Controlled by weight $W$ .
  - Initial DEM Constraint: Controlled by weight $C$ , keeping the solution close to $z_0$ .
- Joint Optimization: Camera pointing errors are corrected simultaneously to prevent them from being misinterpreted as topography.

C. Parameter Study

The authors conducted a systematic three-stage parameter sweep to identify stable operating regimes for OHRC data:

Stage 1 (Decade Sweep): Tested $W \in \{10^4, 10^3, 10^2, 10^1\}$ to find the noise-free threshold.
Stage 2 (Refinement): Fine-tuned $W$ within the optimal decade.
Stage 3 (Interaction): Varied the initial DEM constraint ( $C$ ) and iteration count ( $N$ ).

3. Key Contributions

Independent Data Source: Demonstrated that SfS applied to a third, independent image acts as a genuine source of new topographic data, unconstrained by stereo baseline limitations.
OHRC-Specific Pipeline: Extended the ASP/ISIS pipeline to handle Chandrayaan-2 OHRC data, enabling sub-meter resolution mapping.
Parameter Characterization: Systematically identified optimal parameters ( $W \approx 50$ , $C = 10^{-3}$ ) for OHRC imagery, balancing detail recovery against noise amplification.
Multi-Site Validation: Validated the method across three distinct lunar environments:
- Vikram Landing Region: General terrain.
- Mons Mouton (South Pole): High-latitude, grazing illumination.
- Cyrillus Crater: A "limiting case" with poor geometric separation.
Limiting Case Analysis: Characterized scenarios where SfS fails (e.g., insufficient angular separation between images or partial footprint coverage).

4. Results

Topographic Enhancement: The method successfully recovered fine-scale crater morphology and meter-scale slope variations previously unresolved in the stereo DEM.
- Slope Statistics: In the Vikram region, the slope standard deviation ( $\sigma_s$ ) increased by 22.9% at optimal parameters, indicating significant relief recovery.
- Visual Validation: Hillshaded DEMs showed clear resolution of small craters and ridges absent in the initial DEM.
South Pole Performance: At Mons Mouton, the technique effectively enhanced steep flanks while maintaining the flat plateau, demonstrating sensitivity to grazing solar incidence angles ( $\theta_i > 82^\circ$ ).
Limiting Cases:
- Cyrillus Crater: When the angular separation ( $\Delta \theta_p$ ) between the stereo pair and the shading image was large ( $\approx 30^\circ$ ) but illumination diversity ( $\Delta \theta_i$ ) was low, SfS performance dropped significantly ( $\Delta \sigma_s < 3\%$ ).
- Footprint Coverage: Areas where the shading image $I_3$ did not overlap with the DEM resulted in the solution relaxing back to the initial DEM, highlighting the need for full footprint verification.

5. Significance and Conclusion

This work establishes a reproducible, scientifically grounded methodology for sub-meter lunar topographic mapping. By treating SfS as a source of new data rather than just a refinement filter, the authors have demonstrated that Chandrayaan-2 OHRC imagery can yield DEMs with higher spatial frequency content than stereo reconstruction alone allows.

Key Takeaways:

Operational Parameters: For OHRC data, $W=50$ , $C=10^{-3}$ , and $N \ge 8$ are recommended.
Geometric Requirements: Successful SfS requires both sufficient illumination diversity ( $\Delta \theta_i$ ) and consistent viewing geometry ( $\Delta \theta_p$ ) between the stereo pair and the shading image.
Future Work: The authors suggest moving toward multi-image SfS and hybrid stereo-photoclinometry (fusing stereo and SfS simultaneously) to further improve robustness and absolute accuracy.

This framework significantly enhances the utility of India's Chandrayaan-2 mission data for future lunar exploration, particularly for identifying hazards in complex terrain and polar regions.