This paper presents a method for digital signature verification utilizing a combination of optical character recognition (OCR) and string similarity metrics, implemented in Python. The proposed method involves preprocessing input images, extracting text, and comparing the text for similarity using the SequenceMatcher algorithm. Additionally, the Structural Similarity Index (SSIM) is employed to assess the visual similarity between the images. This paper presents a method for signature verification without delving much into machine learning or deep learning. Experimental results demonstrate the effectiveness of this approach in verifying digital signatures, offering a robust solution for enhancing security in electronic documents.
Introduction
Digital signatures are essential for ensuring authenticity, integrity, and non-repudiation in electronic documents. Traditional cryptographic verification methods are robust but struggle with visual variations in handwritten signatures. This paper proposes a novel approach using Python that combines Optical Character Recognition (OCR) and string similarity metrics to improve verification accuracy by accounting for natural signature variations.
Methodology:
Image Preprocessing: Enhances image quality through grayscale conversion, Gaussian blur, adaptive thresholding, and resizing.
Text Extraction: Uses EasyOCR to extract text from preprocessed signature images.
Text Comparison: Applies Python’s SequenceMatcher to compute similarity between extracted texts, with a threshold to determine further analysis.
Visual Similarity: If text similarity exceeds the threshold, Structural Similarity Index Measure (SSIM) is used to assess the visual likeness, otherwise, the signature is flagged as fraudulent.
Experimental Setup:
The method was tested on a diverse set of genuine and forged signatures collected from various sources, following preprocessing, text extraction, comparison, and SSIM analysis.
Results:
Genuine signatures had higher text similarity (~0.8) and SSIM (~90%) scores.
The approach may struggle with highly dynamic signatures exhibiting variations in pressure or speed.
Future Work:
Plans to enhance the method by addressing limitations and advancing image processing techniques.
Conclusion
The paper showcases an innovative method of digital signature verification utilizing SSIM, text extraction using OCR, text comparison and image preprocessing in Python. This method demonstrates tremendous potential and a reliable method for signature verification. Experiment outcomes showcase that this method can easily identify legit and fraud signatures which in turn can be used to verify integrity of digital signatures.