Review of Artificial Intelligence Methods in
Handwriting Identification Using CNN-RNN for
Textural Features
Chandan Kumar Sonkar
School of Computer Science and Engineering
Galgotias University Uttar Pradesh
c.sonkar1986@gmail.com
Dr. Vinod Kumar
Computer science and engineering
Galgotias University, Greater Noida (U.P.)
vinod242306@gmail.com
Abstract - This review discusses the developments in
handwriting recognition and optical character recognition
(OCR) systems, with a focus on the transformative role of
artificial intelligence (AI) in enhancing accuracy, precision, and
scalability. Modern methods include convolutional neural
networks (CNNs), recurrent neural networks (RNNs), hybrid
CNN-RNN frameworks, and transformer-based architectures.
These methods have been evaluated to address a wide range of
tasks including handwritten text recognition (HTR), signature
verification (SV), and multilingual text analysis. The impressive
results have been achieved across a variety of datasets with
99.98%-digit recognition and more than 97% online
handwritten word recognition. This involves high-end models
such as hybrids CNN-RNN and transformer architecture, which
have shown their effectiveness in capturing spatial and temporal
features and adapting to handwriting variability in writing and
linguistic contexts. The review underlines the importance of
texture and structural feature analysis to progress with AI-
based handwriting systems while emphasizing the extraction of
fine-grained details such as pen pressure and stroke dynamics.
Despite these advances, the issue of managing handwriting
variability, dynamic styles, and underrepresented scripts like
non-Latin languages still persists. Some open areas of research
include robustness improvement, fine-tuning models to better
adapt to variability, and interdisciplinary applications in
forensic analysis, education, and medical diagnostics.
Keywords— Handwriting identification, handwriting
verification, artificial intelligence, CNN-RNN integration,
textural feature analysis, signature verification,
handwritten text recognition, multi-scale analysis,
biometric authentication.
I. INTRODUCTION
Biometric recognition systems, designed to support both
identification and verification, employ two types of traits:
physical traits, such as fingerprints, DNA, and iris patterns,
where the characteristics are relatively stable, and behavioural
traits like signatures, gait, and voice, whose characteristic is
likely to vary in time and can easily be forged, making
precision biometric systems not feasible [1–3]. Among
behavioural traits, handwriting is a unique identifier and is
sometimes referred to as "brain writing" because it reflects an
individual's mood, personality, and motor skills, such as age
and nationality [4–5]. The uniqueness of handwriting makes it
a very important biometric in forensic analysis, security, and
even disease prediction [6]. However, handwriting analysis
remains a difficult problem in artificial intelligence, especially
in Handwritten Text Recognition (HTR) and Signature
Verification (SV). Document Image Analysis is a key area,
which is further subdivided into off-line and on-line systems.
The former is much more complex because dynamic data such
as writing speed and pen pressure are absent. Applications
such as HTR, which converts handwritten documents into
symbolic representations, and SV, which authenticates
signatures, point out the high interpersonal and intrapersonal
variability in handwriting, which makes analysis even more
difficult [7–8]. This is necessary in improving accuracy as
sources of errors could be those as depicted in Fig. 1-
misaligned strokes, unusual styles, among others.
Fig. 1. Examples for sources of error with resulting incorrect embeddings:
a, b unaligned delayed strokes; c unusual writing style; d incorrect type
detection. The original writings are given in the first column [59]
Current AI has fundamentally transformed handwriting
analysis so that identification and verification can be
accelerated further. Starting from the pace and accuracy levels
much quicker than in manual examination; thus, a much faster
and scalable process than what is possible with manual
examination, which is extremely time-consuming and error-
prone. As AI exploits the superiority of machine and deep
learning techniques, such as Convolutional Neural Networks
and Recurrent Neural Networks, complex features are
automatically extracted to improve HTR and SV tasks [9]. AI
systems now are able to differentiate between different forms
of handwriting and cope with vagaries in both printed and
hand-written documents and pictures, which traditional
methods found very difficult to handle. AI-based hand-writing
verification systems have serious practical applications for
forensic investigation, banking, and security and provide
greater precision and less numbers of false positive in writing
authentication [10]. This study briefly outlines the role of AI
in handwriting recognition, focusing on the integration of
648
2025 International Conference on Cognitive Computing in Engineering, Communications, Sciences and Biomedical Health Informatics (IC3ECSBHI)
979-8-3315-1852-3/25/$31.00 ©2025 IEEE
2025 International Conference on Cognitive Computing in Engineering, Communications, Sciences and Biomedical Health Informatics (IC3ECSBHI) | 979-8-3315-1852-3/25/$31.00 ©2025 IEEE | DOI: 10.1109/IC3ECSBHI63591.2025.10990766
Authorized licensed use limited to: University of Delhi. Downloaded on May 26,2025 at 12:28:18 UTC from IEEE Xplore. Restrictions apply.