How ashutoshraj/signaturerecognition works
Project type and purpose
• This is a command-line image-processing pipeline (a small full-stack of modules) for signature verification without deep learning.
• Purpose: take two scanned signature images, preprocess & segment strokes, compute orientation & frequency maps, enhance via Gabor filters, extract keypoints (ORB), and match them to decide if both signatures belong to the same person.
File structure
• Top level:
• main-app.py (entry point / CLI)
• Libraries/ (contains one subpackage per processing stage)
• README.md, LICENSE, config (_config.yml), __init__.py
• Libraries/ modules follow a pip-style package:
• Preprocessing/Preprocessing.py
• StrokeSegmentation/StrokeSegmentation.py
• StrokeOrient/StrokeOrient.py
• StrokeFrequency/StrokeFrequency.py
• StrokeFilter/StrokeFilter.py
• ExtractKeypoints/ExtractKeypoints.py
• No web frontend or database—just a sequential pipeline invoked from the CLI.
README for additional insights
• Steps described in “Technical Details”: CLAHE → grayscale → normalize → stroke segmentation → orientation → frequency estimation → Gabor filtering → thresholding & normalization → thinning & denoising → corner/curve feature extraction → ORB descriptor → brute-force matching → decision by threshold.
• External dependencies: numpy, scipy, matplotlib, OpenCV, scikit-image.