Semi-Automatic Extraction and Comparison of Signature Stroke Paths

Note: The thesis is in collaboration with Forensisches Institut Zürich.

Overview

In forensic handwriting examination, comparisons are typically based on the visual inspection of handwritten samples. However, representing signatures as vector-based stroke paths could enable additional quantitative analyses of writing shapes.

Signatures on paper can be digitised as images, but extracting the underlying writing trajectories remains challenging. Converting handwriting into SVG paths allows strokes to be represented as sequences of coordinates.

These coordinates can subsequently be transformed into x-y series, enabling the application of similarity measures such as Dynamic Time Warping (DTW) or related algorithms for shape comparison.

Figure 1. Example of a DTW-alignment between two signatures.

Figure 1 illustrates an example of a DTW alignment between two signatures.

This master’s thesis investigates methods to convert paper-based signatures into SVG stroke paths and explores algorithms for automatic form comparison of handwriting shapes.

Objectives

Review existing approaches for stroke extraction from handwritten images
Develop or evaluate tools that convert signatures into SVG path representations
Investigate semi-automatic workflows that allow experts to refine extracted stroke paths
Transform SVG paths into coordinate sequences suitable for algorithmic comparison
Evaluate similarity measures such as Dynamic Time Warping (DTW) and related methods
Assess the usefulness of such tools as support systems for forensic handwriting examiners

Methodology

Literature Review: Survey methods for stroke extraction, vectorisation, and handwriting comparison
Stroke Path Extraction: Implement or evaluate algorithms that convert scanned signatures into SVG paths
Semi-Automatic Refinement: Explore workflows where experts can correct or refine extracted stroke paths
Coordinate Transformation: Convert SVG paths into ordered x–y coordinate sequences
Shape Comparison: Apply algorithms such as DTW or other approaches to measure shape similarity
Evaluation: Assess the usefulness and limitations of the proposed workflow
Reporting: Document results in the master’s thesis and final presentation

Skills Required

Programming skills
Basic knowledge of statistics
Interest in AI and handwriting examination

Created: 02.04.2026