Volume 8, Issue 6, June 2023 International Journal of Innovative Science and Research Technology

ISSN No:-2456-2165

Biometric Identification for Recognizing Text

Usingvarious AI Algorithm
Abhishek Jha1, Dr. Hitesh Singh2, Dr. Vivek Kumar3, Dr. Kumud Saxena4
1
(Computer Science and Engineering, NIET, AKTU, India)
2
(Professor, Computer Science and Engineering, NIET, AKTU, India)
3
(Associate Professor, Computer Science and Engineering, NIET, AKTU, India)
4
(HOD, Computer Science and Engineering, NIET, AKTU, India)

Abstract:- Biometric identification has emerged as a wear and tear. It is possible that the original document was
powerful tool for recognizing individuals in various of poor quality because to the changes in toner density that
applications, ranging from security systems to text were present in it [3]. There is a possibility that the scanning
analysis. This paper focuses on the application of process will miss some of the fainter sections that were
biometric identification for recognizing text using a present in the original document, which might result in the
combination of one-hot encoding and convolutional corruption of many characters in the original text.
neural networks (CNNs) through artificial intelligence Gradations in the text picture can be brought about by
(AI). The one-hot encoding technique is employed to scanning on low-quality paper or by printing of low-quality
represent textual data, where each word or character is copies. The performance of optical character recognition is
converted into a binary vector of zeros and ones. This also significantly impacted by other factors, including
representation preserves the unique characteristics of accurate line and word segmentation [4].
the text and enables efficient processing by the CNN
model. The CNN architecture is utilized to learn The method of selecting and extracting the most
meaningful features from the encoded text, capturing unique characteristics from an image is known as feature
important patterns and structures. The integration of AI extraction and selection. This peculiar 5 feature has the
techniques further enhances the accuracy and efficiency significant ability to accentuate or expand the difference
of the biometric identification system. AI algorithms between different class patterns while maintaining its
allow for the automatic extraction of relevant features, invariance to the same class patterns of other kinds. This is
reducing the need for manual feature engineering. The one of its important characteristics [5].
trained CNN model is capable of recognizing text
patterns with a high degree of accuracy, enabling the The class of methods that are used for character
identification of individuals based on their unique recognition in practice is, in principle, not dissimilar to the
textual attributes. Experimental results demonstrate the class of methods that are used for any broad pattern
effectiveness of the proposed approach. The combination recognition issue. On the other hand, based on the
of one-hot encoding and CNN via AI achieves notable characteristics that are employed, the various methodologies
improvements in text recognition performance, for character recognition may be roughly categorized as
follows: Approaches for matching templates and performing
surpassing traditional methods. The system proves
robust to variations in text content, font styles, and sizes, correlations, as well as techniques for analyzing and
highlighting its potential for real-world applications. matching features [6].

Keywords:- CNN, AI, ONE-HOT ENCODING. Template matching and correlation techniques: These
approaches are high-level machine vision algorithms that
I. INTRODUCTION identify the entirety or a portion of an image that match a
standard prototype template. These techniques may be used
In today's world, optical character recognition (OCR) to either a single picture or several images simultaneously.
technologies are utilised for a wide variety of tasks, some of In this step, the pixels of an input character are compared,
which include scanner-based data entry, bank checks, one by one, with character prototypes that have been saved
business cards, automatic mail sorting, handheld price label before [7].
scanners, and a variety of document recognition applications
[1]. These days, the market also contains the commercial Feature analysis and matching techniques: At the
character identification scheme that are accessible to present time, the methods of character recognition that are
purchase. As a direct result of this innovation, researchers utilized most frequently include feature analysis and
are now able to work on the issue of online handwriting matching algorithms. This strategy is also sometimes
recognition. This was made possible as a result of the referred to as the structural analysis method. These
development of electronic tablets that collected the x, y approaches replicate human thinking better than template
coordinate data as well as the movement of the pen tip [2]. matching did, which was the previous standard. Using this
methodology, relevant characteristics are first retrieved from
The quality of the source document that is supplied the input character, and then the extracted features are
into any OCR system has a noteworthy influence on the compared with the feature descriptions of the trained
performance of that system. It is possible that the original characters. Recognition can be achieved by using the
document is rather old and that it's undergone some physical description that is the closest match [8].

IJISRT23JUN2470 www.ijisrt.com 3464

Volume 8, Issue 6, June 2023 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165
Many of today's OCR systems are modelled based on features serially. An entropy fitness function is used for the
mathematical models in order to reduce the amount of selection process. An ensemble classifier is then used to
categorization errors that occur. A method for recognizing label the completed features. This paper presents a method
characters like these might make use of either structural or that can be used to make printed text machine-readable,
pixel-based information. The following list includes some of which has applications in areas such as licence plate
them. Using hyper surfaces in multi-dimensional feature recognition.
spaces, discriminant function classifiers attempt to reduce
the mean-squared classification error by separating the Nadeem and Rizvi [16] proposed using template
feature descriptions of characters that belong to various matching to identify both typed and handwritten characters.
semantic classes [9]. Bayesian classifiers make use of Rather than manually recognising each input pattern, one of
probability theory in order to minimize a loss function that the goals is to create a system that can automatically
is connected with character misclassification. The theories categorise them into the class to which they belong based on
of human and animal vision are the basis for Artificial the information they contain. The system is responsible for
Neural Networks (ANN), which use error BP methods to character recognition. When compared to the recognition
decrease the possibility of incorrect categorization. SVM is rate of typewritten Standard English alphabets fonts
an important innovation in the field of machine learning (94.30%), the rate at which handwritten English alphabets
algorithms. It is a technique that is built on a kernel. There is are understood is significantly lower at 75.42%.
a collection of techniques for supervised learning that may
be used for classification or regression, and these are the Adhvaryu [17] presented an algorithm for matching
techniques [10][11]. templates for alphabets. The letters of the alphabet, from A
to Z, were utilised in tests, and photos were grayscale with
Most recent articles have incorporated a machine Times New Roman font type so that the prototype system
learning method in certain way. Particularly it is utilized could recognise the letters by comparing them. The
widely for the identification of visual characters largely prototype's limitations are that it can only be used to test
owing to the massive dataset availability. The researchers with the alphabet, that it can only use grayscale photographs
frequently utilize a machine learning strategy for a language with the font type, and that it can only use the Template
with big dataset for machine learning in order to study the Matching method to recognise the letters.
meaningful method. This improvement has come with more
computational complexity, as was indicated above, even if Ziaratban et al. [18] suggested a method for character
the frameworks depending on machine learning approaches recognition they called "template matching." In order to do
achieve greater accuracy during classification [12]. There feature extraction, this technique looks for distinct templates
have only been a handful of studies conducted in recent within the input photos. Each template's success rate is
years that have used a traditional method of feature recorded as a feature, and the optimal matching area in an
extraction in conjunction with algorithmic approaches to image is pinpointed and archived.
feature selection and have succeeded in producing state-of- Rajib et al. [19] offer an HMM-based method for
the-art results. Additional research difficulty that requires recognising English handwritten characters. In their
the research community attention is the construction of research, they have used both system-wide and
systems that are able to recognize on-screen text and neighborhood-specific techniques for extracting features.
characters in various circumstances in everyday life The HMM is trained with these features, and trials are run.
situations [13]. They have compiled a database of 13,000 examples, with
II. LITERATURE REVIEW five examples produced for each character by each of 100
writers. After 2600 samples were used to teach the HMM,
A deep learning-based model for ACR has been we are now using the remaining samples to validate the
proposed by Guptha et al. [14]. The data is gathered and recognition architecture. A recognition rate of 98.26% is
then pre-processed with Gaussian filtering and skew achievable with the proposed system.
detection. As an added step, projection profile and
thresholding methods are used to separate the lines and Jamtsho et al. [20] presented a model for a system
characters from the denoised images. Elephant Herding where the system can read aloud the characters/text in the
Optimization (EHO) is used to pick discriminative features photos. A model for OCR and a model for speech synthesis
from the extracted features of the segmented images, while make up the system's two major subsystems. The characters
Enhanced Local Binary Pattern and Inverse Difference and text in the photos are recognised by the OCR model,
Moment Normalized descriptors will be used to extract which then turns them into editable text by a variety of
features from the images. techniques including preprocessing, segmentation, and
classification. Deep learning (DL) neural networks and
When it comes to saving space and maximising machine learning (ML) are used to train both models.
performance, the idea of applying partial least square (PLS)
based feature reduction for optical character recognition was A segmentation-free OCR system that incorporates DL
introduced by Akhtar et al. [15]. Here, we offer a novel techniques, the creation of synthetic training data, and data
approach to automated OCR based on the fusion and augmentation strategies was presented by Namysl and
selection of several attributes from a given set of features. Konya [21]. With the use of enormous text corpora and
Parallel LR (PLS) based selection is used to combine the more than 2000 fonts, we render synthetic training data. We
add geometric distortions to collected samples as well as a

IJISRT23JUN2470 www.ijisrt.com 3465

 Volume 8, Issue 6, June 2023 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165
suggested data augmentation method called alpha- It is worth noting that these approaches can be
compositing with background textures to emulate text combined or augmented with other techniques, such as
appearing in complex natural situations. The CNN encoder attention mechanisms, transfer learning, or ensemble
in our models is used to extract features from text images. methods, to further improve the performance of biometric
identification systems for recognizing text using AI. The
The main focus of Surana et al. [22] was on various choice of approach depends on the specific requirements,
ML techniques that may be used to extract text from available data, and desired level of accuracy.
handwritten documents and photos, recognise them in
digital format, and then translate it in accordance with the IV. COMPARATIVE STUDY OF EXISTING AI
user's needs. ALGORITHMS

An innovative concept and implementation of an Biometric identification for recognizing text using
OCR-based application for Automated NGO Connect using various AI algorithms involves the application of artificial
ML were offered by Sharma et al. [23]. Image de-noising, intelligence techniques to extract and recognize text from
binarization, data extraction, and data conversion are among biometric data. Several approaches can be employed for this
the phases put into practise. Tesseract OCR based on DL is purpose, each with its strengths and limitations. Here, I will
included into the framework along with image processing provide a comparative study of three commonly used
and data visualisation modules. approaches: Optical Character Recognition (OCR),
Convolutional Neural Networks (CNN), and Recurrent
III. DISCUSSION ON VARIOUS APPROACH FOR Neural Networks (RNN).
APPLYING AI TO
BIOMETRICIDENTIFICATION FOR A. Optical Character Recognition (OCR):
RECOGNIZING TEXT OCR is a traditional approach for recognizing text from
images. It involves preprocessing the biometric image to
There are several approaches to biometric enhance text visibility and then applying algorithms to
identification for recognizing text using artificial segment and recognize individual characters. OCR
intelligence (AI). Here are three common approaches: techniques often utilize image processing techniques such as
thresholding, edge detection, and morphological operations.
A. One-Hot Encoding and Machine Learning: Some advantages and limitations of OCR are:
One approach involves using one-hot encoding to
represent text data and then applying machine learning  Advantages:
algorithms for classification or recognition tasks. In this  Well-established technology with mature algorithms.
approach, each word or character is represented by a binary  Effective for recognizing text in structured and well-
vector, where only one element is set to 1, indicating its defined formats.
presence. The encoded text is then fed into machine learning  High accuracy for printed or typed text.
models such as support vector machines (SVM), random
forests, or logistic regression for training and recognition.  Limitations:
 OCR performance may degrade for handwritten or
B. Convolutional Neural Networks (CNN) degraded text.
CNNs have proven to be highly effective in various
 Sensitivity to variations in font style, size, and image
computer vision tasks, including text recognition. In this
quality.
approach, CNNs are used to learn hierarchical
representations of text data. The network typically consists  Limited capability for handling complex layouts or
of convolutional layers that capture local patterns, followed irregular text arrangements.
by pooling layers to reduce dimensionality and fully
connected layers for classification. CNNs excel at capturing B. Convolutional Neural Networks (CNN):
spatial dependencies and extracting meaningful features CNNs are widely used for image recognition tasks,
from text, enabling accurate recognition. including text recognition. They can learn directly from raw
biometric image data by automatically extracting relevant
C. Recurrent Neural Networks (RNN) and Long Short-Term features. CNNs consist of multiple layers of convolutional
Memory (LSTM) and pooling operations to capture local patterns and
RNNs, along with their variant LSTM, are particularly hierarchical representations. Here are some advantages and
suitable for sequential data, such as text. RNNs process text limitations of CNNs:
input step by step, capturing contextual information and
dependencies between words. LSTM, with its ability to  Advantages:
retain long-term dependencies, is well-suited for  Ability to learn and extract complex features directly
understanding the sequential nature of text. This approach from images.
allows for the recognition of text patterns and the  Robust to variations in font style, size, and image
identification of individuals based on their unique textual quality.
attributes.  Effective for recognizing text in different languages and
character sets.

IJISRT23JUN2470 www.ijisrt.com 3466

Volume 8, Issue 6, June 2023 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165
 Limitations:  Longer training times compared to other approaches
 Require large labeled datasets for training, which can be due to sequential processing.
time-consuming and costly to create.
 May struggle with recognizing handwritten or highly In summary, OCR is a well-established approach that
stylized text. performs well on structured text but struggles with
 Limited contextual understanding of the text due to the handwritten or complex text layouts. CNNs excel at
absence of sequential information. extracting features from images and are robust to variations,
while RNNs are suitable for sequential text recognition. The
C. Recurrent Neural Networks (RNN): choice of approach depends on the specific requirements of
RNNs are suitable for processing sequential data, making the biometric identification task and the characteristics of
them a popular choice for tasks involving text recognition. the data being processed.
They can model long-term dependencies by incorporating
V. CHALLENGES AND RESEARCH DIRECTION
feedback connections, allowing them to capture contextual
information. Some advantages and limitations of RNNs are: The use of biometric identification for recognizing text
using various AI algorithms presents both challenges and
 Advantages: opportunities for research. Here are some key challenges
 Ability to capture temporal dependencies in text data. and potential research directions in this field:
 Effective for recognizing handwriting or cursive text. A. Data Collection and Quality:
 Suitable for tasks involving text generation or  Challenge: Collecting a large and diverse dataset of
completion. biometric samples containing text for training and
 Limitations: evaluation can be difficult.
 Research direction: Developing techniques to
 Prone to difficulties in modeling long-term
efficiently collect and annotate biometric text data,
dependencies due to vanishing or exploding gradients.
ensuring its quality, diversity, and representativeness
 Sensitive to noisy or distorted inputs.
across different languages, fonts, and writing styles.

B. Feature

Techniques Training accuracy Testing accuracy

CNN + Adam optimizer 97% 98%
CNN + RMSprop optimizer 99% 99%
CNN+LSTM 92.72% 91%
SVM 96.29% -
VGG-16 model 98.43% 98.16%
LSTM And Adaptive Classifier 98% 98%
Bi-LSTM 96.06% 87.94%
Hybrid PSO-SVM 93% 92.2%

C. Extraction and Representation: E. Cross-Modal Matching and Fusion:

 Challenge: Extracting relevant features from biometric  Challenge: Developing effective methods to match
samples (such as handwriting or fingerprint patterns) to biometric samples with corresponding textual
represent textual information accurately. information accurately.
 Research direction: Investigating advanced feature  Research direction: Investigating cross-modal matching
extraction techniques, including deep learning-based techniques that leverage both the biometric features and
approaches, to capture fine-grained details and unique the associated text, exploring fusion strategies to
characteristics in biometric data that can effectively combine multiple modalities (e.g., fingerprints,
represent text. handwriting, and text) to improve recognition accuracy
and robustness.
D. Algorithm Development:
 Challenge: Designing robust AI algorithms that can F. Generalization and Adaptation:
handle variances in biometric data, including intra-class  Challenge: Ensuring the biometric text recognition
variations and noise. system can generalize well to unseen data and adapt to
 Research direction: Exploring machine learning and different scenarios, such as variations in lighting
deep learning algorithms, such as convolutional neural conditions, device types, or writing instruments.
networks (CNNs), recurrent neural networks (RNNs), or  Research direction: Studying domain adaptation
transformer models, to improve the accuracy and techniques, transfer learning, or unsupervised learning
efficiency of biometric identification for text recognition. approaches to enhance the system's ability to perform
well in diverse real-world settings.

IJISRT23JUN2470 www.ijisrt.com 3467

Volume 8, Issue 6, June 2023 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165
G. Privacy and Security: systems has demonstrated improved accuracy and
 Challenge: Addressing privacy concerns and ensuring robustness compared to traditional methods. The ability of
the security of biometric data during the recognition AI models to learn and recognize complex patterns in text-
process. related biometric characteristics has opened up new avenues
 Research direction: Investigating privacy-preserving for research and practical applications.
techniques, secure encryption methods, and secure
protocols to protect biometric data from unauthorized However, challenges and limitations still exist. The
access or misuse. availability and quality of training data, the need for diverse
and representative datasets, and the interpretability of AI
H. Ethical Considerations: models are ongoing areas of research. Additionally, privacy
 Challenge: Addressing ethical implications related to the and ethical considerations surrounding the collection and
use of biometric identification, such as consent, fairness, use of biometric data must be carefully addressed to ensure
and potential biases. the responsible deployment of these systems.
 Research direction: Examining fairness and bias in
In conclusion, the reviewed literature underscores the
biometric identification algorithms, developing
potential of AI in biometric identification for recognizing
frameworks for ethical use of biometric data, and
text. As technology advances and new research emerges, it
ensuring transparency and accountability in the decision-
is expected that AI-based systems will continue to evolve,
making process.
providing more accurate, efficient, and secure solutions for
I. Real-World Applications: various text-related applications.
 Challenge: Translating research findings into practical
REFERENCES
applications with real-world impact, such as secure
document authentication, forensic investigations, or [1.] Om Prakash Sharma, M. K. Ghose, Krishna Bikram
personalized user experiences. Shah, "An Improved Zone Based Hybrid Feature
 Research direction: Collaborating with industry Extraction Model for Handwritten Alphabets
partners and domain experts to explore specific Recognition Using Euler Number", International
application areas, refining algorithms for scalability, Journal of Soft Computing and Engineering, Vol.2,
usability, and integration into existing systems. Issue 2, pp. 504-58, May 2012.
[2.] C. C. Tappert, C. Y. Suen, and T. Wakahara, ‘‘The
These research directions can help advance the field of state of the artinonline handwriting recognition,’’
biometric identification for recognizing text using AI IEEE Trans. Pattern Anal. Mach.Intell.,vol. 12, no. 8,
algorithms, enabling more accurate and efficient text pp. 787–808, Aug. 1990, doi: 10.1109/34.57669
recognition in various domains and applications. [3.] M. Kumar, S. R. Jindal, M. K. Jindal, and G. S.
Lehal, ‘‘Improvedrecognition results of medieval
VI. CONCLUSION
handwritten Gurmukhi manuscripts usingboosting
In conclusion, the application of artificial intelligence and bagging methodologies,’’ Neural Process. Lett.,
(AI) to biometric identification for recognizing text has vol. 50,pp. 43–56, Sep. 2018.
shown significant promise and potential. Through the [4.] C. Wolf, J.-M. Jolion, and F. Chassaing, ‘‘Text
utilization of various AI algorithms and techniques, localization, enhancementand binarization in
researchers have made substantial progress in developing multimedia documents,’’ in Proc. Object
robust and accurate systems for identifying individuals Recognit.Supported User Interact. Service Robots,
based on their text-related characteristics. vol. 2, 2002, pp. 1037–1040.
[5.] J. Pradeep, E. Srinivasan, and S. Himavathi,
The reviewed literature highlights several key ‘‘Diagonal based featureextraction for handwritten
approaches employed in this field. Handwriting recognition, character recognition system using neuralnetwork,’’
signature verification, voice recognition, natural language in Proc. 3rd Int. Conf. Electron. Comput. Technol.
processing (NLP), keystroke dynamics, and behavioral (ICECT),vol. 4, Apr. 2011, pp. 364–368.
biometrics have all been explored as means to achieve [6.] D. C. Ciresan, U. Meier, L. M. Gambardella, and J.
biometric identification in text-related contexts. Schmidhuber,‘‘Convolutional neural network
committees for handwritten characterclassification,’’
AI algorithms have been trained on large datasets to in Proc. Int. Conf. Document Anal. Recognit., Sep.
recognize and analyze handwritten text, verifying the 2011,pp. 1135–1139.
authenticity of signatures and identifying individual [7.] Pritpal Singh and Sumit Budhiraja, “Feature
speakers based on vocal characteristics. NLP techniques Extraction and Classification Techniques in O.C.R.
have played a crucial role in capturing writing style, Systems for Handwritten Gurmukhi Script”.
linguistic patterns, and vocabulary usage to attribute International Journal of Engineering Research and
authorship, detect plagiarism, and classify text. Keystroke Applications (IJERA), Vol.1, ISSUE 4, pp.1736-
dynamics and behavioral biometrics have allowed for the 1739.
analysis of typing patterns and user behavior, facilitating [8.] N. Suguna and Dr. K. Thanushkodi, “An Improved k-
user authentication and fraud detection.The integration of AI Nearest Neighbour Classification Using Genetic
algorithms and techniques into biometric identification Algorithm”, IJCSI, Vol. 7, Issue 4, No 2, July 2010.

IJISRT23JUN2470 www.ijisrt.com 3468

Volume 8, Issue 6, June 2023 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165
[9.] Emanuel Indermuhle, Marcus Liwicki and Horst
Bunke, “Recognition of Handwritten Historical
Documents: HMM-Adaptation vs. Writer Specific
Training”.
[10.] B.V.Dhandra , Gururaj Mukarambi and Mallikarjun
Hangarge, “Handwritten Kannada Vowels and
English Character Recognition System”, International
Journal of Image Processing and Vision Sciences,Vol.
1 Issue1 ,2012.

IJISRT23JUN2470 www.ijisrt.com 3469