Multi-modal, Sun angle and scale invariant image correspondence using Chandrayaan-2 optical images (OHRC, TMC and IIRS)
“ISRO needs a smart software tool to match and align different types of lunar photos taken by Chandrayaan-2 and other spacecraft. Even if the photos have different lighting, angles, or zoom levels, the software must stitch and align them perfectly down to sub-pixel precision.”
Deconstructed Problem Blueprint
WHAT TO BUILD?
A robust, multi-modal image registration software platform that ingests optical lunar datasets (OHRC, TMC, IIRS) alongside reference frames (LRO NAC, SELENE), detects key features, and maps them with sub-pixel accuracy despite extreme illumination, viewpoint, and scale variations.
WHO IS IT FOR?
ISRO scientists, planetary researchers, mission planning teams, and payload data processing cells at SAC (Space Applications Centre) and ISRO Telemetry, Tracking and Command Network (ISTRAC).
WHY DOES IT EXIST?
Traditional feature-matching algorithms (like SIFT, ORB) fail on lunar surfaces due to lack of atmosphere, harsh shadow variations caused by low solar elevation angles, and massive discrepancies in spatial resolution and sensors between different orbiters.
INPUT → PROCESS → OUTPUT ARCHITECTURE PIPELINE
Chandrayaan-2 optical imagery (OHRC, TMC-2, IIRS) and reference orbital images (LRO NAC, SELENE) in standard formats (GeoTIFF, PDS4, IMG).
Deep learning-based robust feature extraction (e.g., SuperGlue, LightGlue, or invariant self-supervised CNN keypoint descriptors), illumination normalization preprocessing, multi-scale pyramid matching, geometric transformation modeling, and sub-pixel refinement via phase correlation or optical flow.
Geometrically aligned source images matching the reference coordinate system, corresponding match-point coordinate files, and real-time evaluation performance reports.
CONSTRAINTS
- Achieve true sub-pixel accuracy (< 1 pixel registration error).
- Maintain uniform spatial distribution of match points across the entire image tile.
- Handle large raster files (Gigapixel images from OHRC) efficiently with low memory footprints.
- Robust against extreme sun angle and shadow length variations.
JURY METRICS
- Root Mean Square Error (RMSE) of matched tie points.
- Inlier match count and Inlier ratio percentage after RANSAC filtering.
- Uniform spatial distribution score (entropy or quad-tree density balance of match points).
- Processing execution time per gigapixel frame.
CORE BOTTLENECK
Handling severe illumination changes where shadows drastically alter crater and boulder appearances, causing classical and standard deep feature matchers to produce false positives or zero matches.
Why You Should Choose This PS (Pros)
- High prestige space-tech problem statement directly backed by ISRO.
- Clear, quantifiable mathematical and algorithmic deliverables (RMSE, inlier counts).
- Great potential for creating reusable open-source planetary science software modules.
Challenges & Risks to Watch (Cons)
- Heavy computation required for high-resolution OHRC imagery needs optimization.
- Handling massive raster data formats (GeoTIFF/PDS) can crash unoptimized backend pipelines during evaluation.
- Lack of immediate ground-truth pairs without preprocessing test datasets.
Recommended Tech Stack & Grand Finale Features
Killer Features to Impress the SIH Jury
View Official SIH 2026 Portal Raw Text▼
Background Image Registration is the process of aligning two or more images of the same scene taken at different times, from different viewpoints, or by different sensors into a common coordinate system.
It has two main components:
Description The process of lunar images registration involves finding match points between source and reference image and then aligning the source image with the reference image. The key challenges involved in this process are as follows:
Expected Solution Generic software solution for finding correspondence between Chandrayaan-2 acquired optical images and Lunar reference images with a sub-pixel accuracy of source image maintaining uniform distribution across the images.