Papers
Topics
Authors
Recent
Search
2000 character limit reached

MarkMatch: Same-Hand Stuffing Detection

Published 11 May 2025 in cs.CV | (2505.07032v1)

Abstract: We present MarkMatch, a retrieval system for detecting whether two paper ballot marks were filled by the same hand. Unlike the previous SOTA method BubbleSig, which used binary classification on isolated mark pairs, MarkMatch ranks stylistic similarity between a query mark and a mark in the database using contrastive learning. Our model is trained with a dense batch similarity matrix and a dual loss objective. Each sample is contrasted against many negatives within each batch, enabling the model to learn subtle handwriting difference and improve generalization under handwriting variation and visual noise, while diagonal supervision reinforces high confidence on true matches. The model achieves an F1 score of 0.943, surpassing BubbleSig's best performance. MarkMatch also integrates Segment Anything Model for flexible mark extraction via box- or point-based prompts. The system offers election auditors a practical tool for visual, non-biometric investigation of suspicious ballots.

Summary

A Review of MarkMatch: Same-Hand Stuffing Detection

The research paper introduces MarkMatch, an advanced retrieval system specifically designed for detecting instances of same-hand stuffing in hand-marked paper ballots. This system marks a significant improvement over existing state-of-the-art methods, such as BubbleSig, by employing a contrastive learning-based approach to assess stylistic similarities between ballot marks drawn by potentially the same individual.

Model Design and Methodology

MarkMatch addresses distinct shortcomings observed in prior methods like BubbleSig, which primarily relied on binary classification of isolated mark pairs. Instead, MarkMatch ranks candidate marks from a pool based on their similarity to a query mark. This approach utilizes a dense similarity matrix and a dual loss objective for training, contrasting each sample against multiple negatives within the batch. This helps the model learn subtle handwriting differences and enhances its generalization capacity despite inherent handwriting variations and visual noise. Importantly, diagonal supervision is used to reinforce high confidence on true matches, providing robustness against intra-writer variability and ambiguous cases. The paper reports that this model achieves an impressive F1 score of 0.943, significantly surpassing BubbleSig’s best performance.

System Integration

MarkMatch integrates the Segment Anything Model (SAM) to offer flexible, prompt-based mark segmentation using bounding boxes or point clicks. This feature improves upon traditional methods which often depended on time-consuming manual alignment techniques. The integration allows for efficient extraction of mark segments directly from ballots, enabling users to query marks against a database swiftly. This capability gives election auditors a practical, visual tool for non-biometric analysis of suspicious ballots.

Demonstration and Practical Features

The interactive platform provided by MarkMatch facilitates detailed analysis of hand-drawn ballot marks through visual similarity assessments. The system incorporates features such as a ranking table and heatmap visualization for each query mark, displaying retrieval confidence distributions across mark comparisons. These components help users evaluate both statistical evidence and visual resemblance of suspicious marks, lending credence to the model’s predictions.

Implications and Future Work

Practically, MarkMatch serves as an effective non-biometric tool, enhancing transparency and scalability in ballot reviews. Theoretically, this model introduces novel ideas into the domain of contrastive learning, demonstrating how it can be applied to handwriting detection tasks. Future developments may focus on extending the methodology to broader applications within the field of AI, including more sophisticated learning models with enhanced noise resistance and improved segmentation techniques. Researchers may also explore the expansion of MarkMatch’s capabilities beyond election security, into other domains requiring precise detection of stylistic similarities or verification tasks.

Overall, MarkMatch exemplifies a significant advancement in the detection of ballot stuffing and presents a robust framework for enhancing election security through deep learning methodologies.

Paper to Video (Beta)

No one has generated a video about this paper yet.

Whiteboard

No one has generated a whiteboard explanation for this paper yet.

Open Problems

We haven't generated a list of open problems mentioned in this paper yet.

Continue Learning

We haven't generated follow-up questions for this paper yet.

Collections

Sign up for free to add this paper to one or more collections.