New COLD Feature Based Handwriting Analysis for Enthnicity/Nationality Identification

Sauradip Nag, Palaiahnakote Shivakumara, Yirui Wu, Umapada Pal, Tong Lu

TL;DR

This paper introduces a novel method for identifying a writer's nationality from their English handwriting by analyzing the "Cloud of Line Distribution" (COLD) of character strokes. By extracting features based on the shape of these COLD distributions, the method achieves promising results in distinguishing between different nationalities, offering a new tool for forensic applications.

What Problem Does This Paper Address?

Identifying individuals based on their handwriting is a long-standing area of research, often used in forensic investigations. While traditional methods focus on writer identification, this paper tackles a more specific challenge: ethnicity or nationality identification from handwriting. This is particularly relevant in forensic scenarios where crimes might involve individuals from different national backgrounds. The authors highlight that existing biometric methods can be sensitive to environmental factors and computationally expensive. Handwriting analysis offers a potential alternative, but accurately identifying nationality is difficult due to the inherent variations in handwriting influenced by factors like ink, pen, paper, age, gender, and individual differences. The paper aims to bridge this gap by proposing a method that leverages subtle, yet consistent, differences in writing styles across different nationalities.

Methods

The proposed method, dubbed COLD Feature based Handwriting Analysis, involves several key steps:

  1. Text Line Segmentation: The process begins by segmenting individual text lines from the document. This is achieved using a combination of horizontal projection profile analysis and tangent angle calculations, effectively identifying text regions and distinguishing them from potential baseline artifacts.
  2. Rule Line Removal: To ensure clean text line analysis, the method removes horizontal rule lines that might be present on the paper. This is done by identifying pixels that move parallel to the tangent angle of the text line contours.
  3. Canny Edge Detection: After preprocessing, Canny edge detection is applied to the segmented text lines to extract the contours of character components.
  4. COLD Feature Extraction: This is the core innovation. For each edge component, the method identifies "dominant points" using polygonal approximation. These points are then connected to form line segments. For each segment, the angle and length are calculated, mapping it to a point in a polar domain. Aggregating these points for all line segments within a text line creates the Cloud of Line Distribution (COLD). The authors hypothesize that the shape of this COLD distribution varies predictably with nationality due to differences in national writing styles (e.g., more cursive vs. more straight strokes).
  5. Feature Engineering using Principal Axis: To make the COLD distribution robust to variations, the principal axis of the distribution is estimated using Principal Component Analysis (PCA). Distances from pixels to this principal axis are then calculated (LD and RD), and their absolute difference forms the feature vector. This approach aims to capture the overall shape and orientation of the COLD distribution.
  6. Classification: Finally, a Support Vector Machine (SVM) classifier, specifically a Gaussian Kernel with a one-vs-one scheme, is used to classify the extracted feature vectors and predict the nationality.

Datasets

The paper introduces a newly created dataset for evaluating their method. This dataset consists of 500 English handwritten text line images from five Asian countries: Bangladesh, China, India, Malaysia, and Iran. The images were collected from individuals aged 20-25, who were free to choose their writing tools and topics. The dataset is described as complex, containing low-contrast, blurred, and ruled/unruled page images, making it suitable for robust evaluation. The authors note that existing datasets like QUWI [8] are available but do not cover the specific Asian countries of interest.

Key Insights and Ideas

  • COLD as a Nationality Fingerprint: The central idea is that the "Cloud of Line Distribution" of character strokes, when analyzed in polar coordinates, forms a unique "shape" that is characteristic of a writer's nationality. This shape is influenced by inherent differences in national writing styles.
  • Principal Axis for Robustness: Using the principal axis of the COLD distribution as a reference for feature extraction makes the method invariant to rotation and scaling, and more robust to minor pixel variations.
  • Bridging the Gap in Forensic Applications: The paper addresses a practical need in forensic science by providing a method to identify potential nationalities of individuals involved in crimes, even when only their handwriting is available.
  • Focus on Asian Countries: The choice of Asian countries is motivated by their shared cultural influences and diverse writing systems, which are hypothesized to lead to distinct yet recognizable English handwriting patterns.

Key Results

The proposed method demonstrates promising performance:

  • Classification Rate: The paper reports a 75% classification rate for the proposed method on their custom dataset.
  • Outperforms Existing Method: The proposed method significantly outperforms an existing geometric feature-based method [6], achieving a 75% CR compared to the existing method's 38.6% CR. This highlights the effectiveness of the COLD features.
  • Robustness: The COLD features are shown to be robust to intra-variations (different writers from the same nation) and somewhat invariant to rotation, scaling, and distortion.

Limitations and Future Work

  • Dataset Size and Scope: While the authors created a new dataset, it's limited to 500 text lines from 5 Asian countries. The performance on a larger, more diverse dataset with more nationalities and writers would be valuable.
  • Extension to Other Nations: The authors believe the proposed method can be extended to other nations due to the robustness and generic nature of the features.
  • Multi-lingual and Multi-script Identification: A future direction mentioned is extending the idea to identify provinces within multi-lingual countries (like India, Singapore, Malaysia) that might use multiple scripts. This would involve analyzing different scripts within a single country.

Summary updated

Abstract

Identifying crime for forensic investigating teams when crimes involve people of different nationals is challenging. This paper proposes a new method for ethnicity (nationality) identification based on Cloud of Line Distribution (COLD) features of handwriting components. The proposed method, at first, uses tangent angle of the contour pixels in each row and the mean of intensity values of each row for segmenting text lines. For segmented text lines, we use tangent angle and direction of base lines to remove rule lines in the image. We use polygonal approximation for finding dominant points for contours of edge components. Then the proposed method connects the nearest dominant points of every dominant point, which results in line segments of dominant point pairs. For each line segment, the proposed method estimates angle and length, which gives a point in polar domain. For all the line segments, the proposed method generates dense points in polar domain, which results in COLD distribution. As character component shapes change, according to nationals, the shape of the distribution changes. This observation is extracted based on distance from pixels of distribution to Principal Axis of the distribution. Then the features are subjected to an SVM classifier for identifying nationals. Experiments are conducted on a complex dataset, which show the proposed method is effective and outperforms the existing method.
BibTeX
@inproceedings{nag2018new,
  author = {Sauradip Nag and Palaiahnakote Shivakumara and Yirui Wu and Umapada Pal and Tong Lu},
  title = {{New COLD Feature Based Handwriting Analysis for Enthnicity/Nationality Identification}},
  booktitle = {ICFHR},
  year = {2018},
  url = {https://openreview.net/forum?id=HneDU8QWHB3},
}
Figure