IIIT-Bangalore Builds Single-Image Visual Defect Detector
Researchers at IIIT-Bangalore built a computer-vision tool that inspects factory products by comparing each item to a single 'golden' reference image, showcased at the Bengaluru Tech Summit 2025 and published Nov. 29, 2025. The system uses ECC alignment, CLAHE and a learned baseline noise mask to correct misalignment and filter reflections, claiming up to 98% accuracy and processing images in under 13 seconds on a CPU. It requires no GPU or retraining.
Key Points
- 1Develops single-reference vision system detecting hairline scratches, dents, alignment errors using ECC alignment and CLAHE.
- 2Reduces dependency on large labeled datasets and GPUs, avoiding expensive retraining for small manufacturers.
- 3Enables sub-pixel defect detection at up to 98% accuracy, running under 13 seconds per image on CPU.
Scoring Rationale
Validated, low-cost single-image inspection delivers high practical value for SMEs; limited novelty beyond established alignment and filtering techniques.
Sources
Public references used for this report.
Practice interview problems based on real data
1,625 SQL & Python problems across 15 industry datasets — the exact type of data you work with.
Try 250 free problems
