Handwriting Biometrics: Applications and Future Trends in e-Security and e-Health

The Biometric Data Behind Handwriting

• Online devices record time-stamped pen or fingertip movement, including position, surface contact, pressure, orientation, and timing.Derived features can include velocity, duration, width, and height.Source: The Biometric Data Behind Handwriting · p. 4
• Some tablets track in-air coordinates only while the pen is within about 1 cm; timestamps retain in-air duration, unlike classic offline scanning.Offline methods also lose or make timing difficult to recover.Source: The Biometric Data Behind Handwriting · p. 4
• 3D in-air tracking can use camera-based MOCAP or IMUs at around 100 Hz, with occlusion and accelerometer drift as respective drawbacks.Camera systems require a fixed space, while IMU systems are portable.Source: The Biometric Data Behind Handwriting · p. 4
• While 2D models and algorithms are established, the potential of 3D handwriting systems remains unexplored.The paper identifies 3D acquisition as a promising research line.Source: The Biometric Data Behind Handwriting · pp. 4–5

Security Applications

• Handwriting recognition studies include signature-based verification and recognition, with less common text-dependent or text-independent approaches,, [109], [15].Competitions including BSEC, SVC, and SigWIComp summarize the state of the art.Source: Security Applications · p. 5
• Online studies report identification rates of 95% for five English words and 92.38% for one capitalized word, rising to 96.46% with two words.The two-word result also reports a minimum detection cost function of 0.033 (3.3%).Source: Security Applications · p. 5
• Fingertip-written smartphone passwords yielded verification rates of 96.2% for seven digits and 94.1% for four digits.A follow-up considered characters and symbols in cross-device, in-the-wild conditions across 94 smartphone models.Source: Security Applications · p. 5
• Graphical inputs such as drawings usually perform worse than natural handwriting or signatures, though they may suit low-security uses or serve as a second factor,,.Source: Security Applications · p. 5
• Forensic challenges include identifying a forger and detecting disguising, in which writers alter their habits; online timing information may help future efforts,.Disguising may be intended to support later denial of authorship.Source: Security Applications · pp. 5–6

Health Applications

• Adding in-air features raised Parkinsonian dysgraphia classification to 84%, 6% above on-surface features alone, and combining both yielded 86%, [80].In-air and on-surface movements contain approximately equal, notably non-redundant information.Source: Health Applications · p. 6
• In-air duration was seven times longer for Alzheimer’s disease patients than controls, versus three times longer on-surface.In-air measures have also been studied for autism and developmental dysgraphia; jerk features discriminated dysgraphia from controls,, [108], [106].Source: Health Applications · p. 6
• Kinematic models, including the Kinematic Theory and its Delta and Sigma–Lognormal models, have been used to represent human movement and study fine motricity, [92], [93],.Source: Health Applications · p. 6
• The Sigma-Lognormal model reflects neuromuscular features that are difficult to impersonate; its lognormality principle links controlled movement with lognormal velocity profiles,,.The cited account associates development and aging or health issues with movement profiles moving toward or away from lognormality, respectively.Source: Health Applications · p. 7

Metadata Applications

• Handwriting metadata includes technical sensor characteristics and personal attributes such as language, dialect, origin, gender, and age,, [20].Technical metadata helps ensure minimum acceptable quality of raw biometric signals.Source: Metadata Applications · p. 7
• Gender classification using four repetitions of one uppercase word averaged 68%, rising to 72.6% with sixteen words; combining in-air and on-surface information yielded 74%.The authors note that chance accuracy in this two-class problem would be 50%.Source: Metadata Applications · p. 7
• Reported demographic classification accuracies include 77.5% for gender, 86.6% for age, and 74.4% for handedness using measurements such as pressure and stroke formation.A separate study reported 67.06% for gender and 84.66% for handedness.Source: Metadata Applications · p. 7
• A tablet-acquired database links handwriting and drawing tasks to questionnaire-assessed anxiety, stress, and depression; anxiety and stress recognition outperformed depression recognition.The database includes 129 participants and is freely distributed.Source: Metadata Applications · pp. 7–8

Why joint consideration matters

• Security and health applications are often treated separately, although the paper argues they cannot always be isolated.The Personnal Digital Bodyguard vision combines sensitive-data protection with signature and writer verification.Source: Applications Combining Security and… · p. 8

Authentication beyond identity

• Security systems may need to account for context beyond identity, including stress that could indicate coercion and health conditions affecting a person’s understanding.The paper also notes research on alcohol and caffeine effects on handwriting.Source: Applications Combining Security and… · pp. 8–9
• A genuine signature may not establish legal validity when the signer’s health state affects their understanding of the act.The paper also compares this problem with replay attacks, where a genuine signal recorded earlier is presented again as a fake attempt.Source: Applications Combining Security and… · pp. 8–9

Privacy risks across security and health

• Handwriting samples collected for security can reveal health information without the user’s knowledge, and long-lived samples may later yield information not intended at acquisition.The paper calls for technical privacy protections and legal safeguards against foreseeable misuse.Source: Applications Combining Security and… · p. 9
• Anonymizing handwriting datasets for health research may fail if the samples still permit biometric identification.The paper notes this concern for both handwriting and drawings.Source: Applications Combining Security and… · p. 9

Current research landscape

• The paper describes signature-based security as mature, while handwritten-text and drawing security and several health applications remain incipient.Source: Future Trends · p. 10

Research priorities

• The proposed research agenda spans handwriting processing and recognition, health diagnosis and monitoring, and removing unwanted health or identity information from handwriting representations.These are presented as four broad areas, including both directions of privacy protection.Source: Future Trends · p. 10

Security applications

• Security research should develop handwritten-text author identification in text-dependent and text-independent modes, including whether writers are coerced or free to choose the text.The paper distinguishes this goal from classical optical character recognition.Source: Future Trends · p. 11
• Drawing-based author identification is a proposed security direction, while combining signature and text or testing crossed recognition remains underexplored.The paper notes that drawing-based smartphone unlock possibilities are more limited than handwritten drawings.Source: Future Trends · p. 11

Health applications

• Health research directions include using handwritten text and drawings for healthy-aging assessment and drug-effect evaluation, and investigating signatures for disease diagnosis.The paper notes that signatures may have health potential despite tending to involve little cognitive effort.Source: Future Trends · p. 11

Joint security and health applications

• The agenda includes detecting stress from handwriting for security applications such as coercion detection, as well as preliminary handwriting tests for substance abuse.The paper describes such tests as potentially cheap and non-invasive.Source: Future Trends · p. 11

Privacy

• Privacy-preserving algorithms should hide identity in samples used for diagnosis and hide health information in samples used for identification.Source: Future Trends · p. 11

Research resources

• Research datasets should cover multiple acquisition sessions and diverse tasks, with ethical committee approval, to address temporal user variability.Suggested tasks include cursive text, capitals and digits, and drawings.Source: Future Trends · p. 11

• Figure 1 presents on-surface/in-air properties for biometric handwriting.Source: Figure 1 · p. 19
• Figure 2 shows on-line handwriting from healthy and unhealthy subjects, with in-air traces in red and on-surface traces in blue.Source: Figure 2 · p. 20
• Figure 3 depicts devices for capturing handwriting information.Source: Figure 3 · p. 21
• Figure 4 presents a direction toward combining identification and health properties of biometric handwriting information.Source: Figure 4 · p. 22

• Table 1 covers Kinematic Theory applications related to handwriting biometrics.Source: Table 1 · p. 23
• Table 2 is titled around future developments in handwriting analysis, covering research areas and specific problems.Source: Table 2 · p. 24
• The section introduces future areas under the heading “Necessity of handwriting data for working on future areas.”Source: TABLES · p. 24