Image Comparison (Function Blocks) for Industrial Automation
Budget: ₹1,500 – ₹12,500 INR
I need a compact computer-vision solution that can tell when two images of industrial-automation function blocks represent the same logic, even if the graphics differ. The input images will be object representations—think ladder-diagram symbols, IEC 61131-3 blocks, or similar—not handwritten text.
The tool must recognize equivalence across:
• Different drawing styles
• Different orientations (rotated or flipped)
• Different sizes or aspect ratios
A typical use-case: an engineer drags two screenshots of schematic pages into the app and instantly learns whether the underlying block is identical, variant, or unrelated, along with a similarity score.
Key expectations
• Accepts common raster formats (PNG, JPEG) and preferably simple SVGs.
• Returns a clear match/no-match flag plus confidence.
• Handles small batches (≤50 images) without cloud dependency; Python with OpenCV, TensorFlow or PyTorch is fine, but I’m open to C++ or another performant stack.
• Accuracy tested on a starter set of block icons I will provide; ≥95 % correct matches under the three variation scenarios above is the acceptance threshold.
Please outline the approach (feature-based, deep learning, hybrid, etc.), estimated timeline, and any similar work you’ve done.
The tool must recognize equivalence across:
• Different drawing styles
• Different orientations (rotated or flipped)
• Different sizes or aspect ratios
A typical use-case: an engineer drags two screenshots of schematic pages into the app and instantly learns whether the underlying block is identical, variant, or unrelated, along with a similarity score.
Key expectations
• Accepts common raster formats (PNG, JPEG) and preferably simple SVGs.
• Returns a clear match/no-match flag plus confidence.
• Handles small batches (≤50 images) without cloud dependency; Python with OpenCV, TensorFlow or PyTorch is fine, but I’m open to C++ or another performant stack.
• Accuracy tested on a starter set of block icons I will provide; ≥95 % correct matches under the three variation scenarios above is the acceptance threshold.
Please outline the approach (feature-based, deep learning, hybrid, etc.), estimated timeline, and any similar work you’ve done.