See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/355030802

Personality Prediction Through Handwriting Analysis Using Convolutional

Neural Networks

Chapter · January 2022

DOI: 10.1007/978-981-16-3802-2_5

CITATIONS READS

13 2,710

2 authors, including:

Gyanendra Chaubey
Indian Institute of Technology Jodhpur
8 PUBLICATIONS 175 CITATIONS

SEE PROFILE

All content following this page was uploaded by Gyanendra Chaubey on 07 October 2021.

The user has requested enhancement of the downloaded file.

Chapter 5
Personality Prediction Through
Handwriting Analysis Using
Convolutional Neural Networks

Gyanendra Chaubey and Siddhartha Kumar Arjaria

1 Introduction

The real personality of a person including his behavior, emotional outlay, self-esteem,
anger, imagination, fears, honesty, and many other personality traits predicted using
handwriting analysis. Graphology or handwriting analysis is described as scien-
tific study and analysis of handwriting. It is a way of identifying, evaluating, and
interpreting behavior of an individual. The performance is measured by examining
multiple samples to identify the writer with a piece of handwriting. Professional
handwriting examiner is called graphologist. Handwriting is also one of the expres-
sive ways that tells about the nature, psychology, and behavior of writer. It is unique to
each individual. Each personality traits has a neurological brain pattern in the human
brain, and each neurological brain pattern design delivers one of a kind neuromus-
cular movement which is same for each person who has those specific personality
traits [1].
Graphology was used always; but nowadays, it has been related to the physical
personality and emotional activities of the handwriters and their current domain
dispute. In the present learning system, graphology is identified by some psycholog-
ical analysis. Graphologists thought that the mind of a human form the characters
based on the attitude of the writer [2]. Handwriting is a complex activity, and it is
considered as an “overloaded” skill including highly sequencing of moments. As it
has been concluded by WEINTRAUB (1997), several theoretical models and struc-
tures that was instructed that handwriting activities concludes retrieving the form,
shape, size, and angle of letters, relating these to be their sounds (phonemes), remem-
bering all the specific parameters, and converting them by main execution to the study
paper [3].

G. Chaubey (B) · S. K. Arjaria

Department of Information Technology, Rajkiya Engineering College, Banda 210201, India

© The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2022 59
R. Tiwari et al. (eds.), Proceedings of International Conference on Computational
Intelligence, Algorithms for Intelligent Systems,
https://doi.org/10.1007/978-981-16-3802-2_5
60 G. Chaubey and S. K. Arjaria

In this paper, architecture of convolutional neural network has been proposed

to predict the personality of a person from the different features extracted from his
handwriting. Handwriting is also called brain writing which determined the reflection
of an individual brain in many ways. The prime features of handwriting are slant of
word, baseline, pen pressure, break in writing, margin, and design of letters [1, 4].
Literally, with the generation or time, handwriting performance becomes auto-
matic and motor technology. More researchers’ studies have shown that adults need
to think mostly about the size, shape, form, and positions of letters, tending to write
slowly and in larger alphabetic letters. On other hand, handwriting is one of the
old models of correspondence in our civilization which has developed over time.
Another way some writer learn to write anything by copying forms and shapes from
a standardized book store which itself increases as a functionality of the geographical
place, temporal, society, and culture preferences [3].
Writer evaluation or identification is used for different purpose like, for monetary
activities, forensic, and security purpose. Handwriting analysis or graphology is also
used in criminal justice system or organization to judge the culprits of handwritten
documents. Handwriting analysis or graphology plays a vital role in the fields such
as medical diagnosis, recruitment, in selection of staff, forensic, and psychology.
Graphology reveals the physiological and psychological conditions of the patient,
so we can say that it is used as an essential tool in psychological and medical
diagnosis [5].

2 Related Works

In the recent works, we have found that there is no standard method has been devel-
oped for predicting behavior of a person on the basis of handwriting. There are many
researchers have been done in the area of computer science to develop such system
which is able to recognize the personality of a person based on handwriting in an
easy way.
In the further paragraphs, the recent works which has been done to determine the
handwriting and psychological traits are described.
Fallah and Khotanlous [6] explained about the personality prediction using hand-
writing studies of an individual. They have used Minisota multi-phasic personality
inventory for training the model and used neural network (NN) to classify the proper-
ties which are not related to the writer and employed Hidden Markov Model (HMM)
to classify the properties of target writer. The accuracy obtained in this model was
over 70%.
A system was proposed to predict the personality traits from handwriting using
various features. The feature includes “The lower loop of letter ‘y,’ letter ‘t’, ” the slant
of writing and pen pressure. The personality traits of the writer on the Myers-Briggs
Type Indicator was classified using a rule-based classifier. The accuracy obtained
over the scale of 70% [7].
5 Personality Prediction Through Handwriting … 61

The combined model of Support Vector Machine (SVM), Adaboost, and K-

NN classifier is used with each of seven personality features for analysis of the
handwriting traits. The final result was ranging from 62.5% to 83.9% [8].
Siddiqui et al. used Artificial Neural Network (ANN) and Support Vector Machine
(SVM) in order to identify male and female from respective handwriting. The features
used to analyze are slant, curvature, legibility, and texture. Based on the database of
males and females handwriting, the system predicted over 80% accuracy on analysis
of gender of writer [9].
Using random forest algorithm and kernel discriminant analysis, a model was
proposed to analyze handwriting on the basis of geometric features. This system
predicts gender, age, nationality with 75.05%, 55.76%, and 53.66% accuracies,
respectively, for an individual writer. But when collective result of all writers was
estimated, then accuracies are gender: 73.59%, age: 60.62% and nationality: 47.98%
when each wrote a different text [10].
Gil Luria and Sara Rosenblum created a model to discriminate between the low
and high mental workloads on the basis of handwriting features. 56 participants were
asked to write three arithmetic progressions on a digitizer. The features extracted
on the basis of temporal spaces but less in pressure space. The three clusters of
the handwriting types were identified using data reduction and result obtained that
mental workload is obtained by affected handwriting behavior [11].
Zaarour et al. used Bayesian-network-based model to enhance the performance
of students by using a model which takes input of different drawing and writings.
The whole model was able to predict the writing behavior of a child which can be
analyzed by a psychologist in order to improve the child’s education [12].
In the similar way, Sudirman et al. worked on prediction of behavior of children
using handwriting, since children have less impact in terms of cultural background,
and their cognition rate is also too good. Hence, children are the best choice in order
to do the analysis of handwriting. That is why they developed an automatic system
which is able to find out the developmental disorders among children. The system
was able to perform with accuracy of over 78% [13].
Another analysis based on handwriting is the detection of deceit in an individual’s
character. Luria et al. worked on this research in which they developed which analyzes
the handwriting in order to detect the false statement provided by a patient about
his health. In their first step, they used a paper to write the true/false statements
about the medical conditions and that paper was further linked to a digitizer. All
the subjects belonging to the experiment were asked to write. In the second step,
the outputs of a particular subject containing deceptive and truthful writing were
compared accordingly, and these subjects are divided into three different groups
based on handwriting profiles. In the further analysis, it was found that the deceptive
writing is much different from other two types of writing. Deceptive writing takes
more time and is broader. The other two types of writings were different in both
spatial and temporal vectors [14].
As we are aware about detecting deceit, human diseases can also be predicted using
handwriting. In this research, a system is proposed which detects diabetic’s disease
using handwriting with accuracy over 80% [15]. Researchers used handwriting in
62 G. Chaubey and S. K. Arjaria

order to predict a disease named micrographia, i.e., this is a disease in which there
is decrease in the size of velocity, acceleration, and size of letters while writing.
This disease is also associated with Parkinson’s disease (PD). It was tested on the
patients who were diagnosed of PD, and the result gave 80% accuracy on 75 tested
subjects [16].
In order to analyze the multitude of handwriting features, there is a wide range
of methods that can be employed. The first step is the normalization of written
sample in the offline handwriting analysis to be ensured for any possible noise.
Morphological methods are typically used for removing background noise as a part
of normalization phase [17]. Laplace filters, unsharp masking, or gradient masking
are used for sharpening in the normalization phase [18]. Contrast enhancement is
an important factor for obtaining the high accuracy in handwriting analysis [19].
Methods of contour smoothing are essential for the written letters to be a part of
normalization phase [20]. Different types of thresholding can be employed after all
processing steps to handwriting sample being compressed and converted to grayscale
image [21]. Through page segmentation, the written text needs to delimit by post
compression where methods for examining the background and foreground region
are employed. The most common was the whitespace rectangles segmentation [22].
Vertical projection profile method has been used to segment the handwriting images
into text lines and words [23]. Different classifiers have been used for the features
classification of handwriting behaviors. In the above sections, we have presented the
analysis.
The percentage of personality traits like pessimistic, optimistic, balanced and shy,
etc., is predicted by Nikit Lemos et al. in 2018 [24] using the convolutional neural
network on the English Handwriting. Fatimah et al. [25] used the CNN method to
evaluate the accuracy of the personality prediction on six different traits on structural-
based approach and symbol-based approach. Nijil Raj et al. [26] used extracted seven
features and then applied CNN mixed with MLP to predict the personality traits using
calligraphy.
Keeping in mind all the above proposed works, this research work provides a
convolutional neural network model for predicting the five important personality
behaviors by analysis of handwriting. This system is proposed to give the accuracy
of the convolutional neural network (CNN) in analyzing the handwriting to predict the
five big personality traits. In next session, we will present our model and architecture
of the system.

3 Research Technique and Design

The five big personality traits which are going to predict here are:
(a) Agreeableness,
(b) Conscientiousness,
(c) Extraversion,
5 Personality Prediction Through Handwriting … 63

(a) English Handwriting

(b) Hindi Handwriting

Fig. 1 Samples of handwriting

(d) Neuroticism,
(e) Openness.
These five big personality traits are treated as target classes for the convolutional
neural networks.
The image dataset of each class has been created by taking the individual hand-
writing of students of an engineering college and taking their personality kinds as
classes. The handwriting of both Hindi and English has been taken. The handwriting
samples contain a paragraph including all the letters of English alphabet as well as
all the letters of Hindi alphabets. The samples of the handwriting are shown below
in Fig.1.

3.1 CNN Architecture for Handwriting Analysis:

A CNN consists of the number of convolution layers and pooling layers. An input
image is p × p × m where p × p is the height and width of the image and m is the
number of channels, for example, an rgb image has 3, and convolution layer has k
filters of size q × q × n where q < pq < p and n ≤ mn ≤ m. The architecture of
the convolutional neural network proposed to solve the problem is shown below in
Fig. 2.
The color coding of the architecture of the convolutional neural network is given
in Fig. 3.
64 G. Chaubey and S. K. Arjaria

Fig. 2 Architecture of convolutional neural network (CNN)

Fig. 3 Color coding for the Input Layer

architecture of CNN
Convolutional Layer
Pooling Layer
Flatten Layer
Dense Layer
Dense Layer1 (Output Layer)

Input Layer: The input layer of the architecture takes the input from the dataset.
Convolution Layer: Convolution layer performs the convolution operation on the
image to extract the features from the image. A weighted matrix is formed using the
input image and the kernel. The convolution operation is given by Eq. (1)
 
Wi ∗ X i Wi ∗ X i (1)

Pooling Layer: Using the features obtained after convolution for classification,
the pooling layer takes small rectangular blocks from the convolutional layer and
subsamples it to produce a single output from that block.
Flatten Layer: In between the pooling layer and fully connected layer, there is a
layer known as Flatten layer. This layer transforms a two-dimensional feature matrix
into a vector that can be fed into a fully connected neural network layer.
Dense Layer: The deeply connected neural network layer is known as dense layer.
It is the most commonly and frequently used layer takes up the input and returns the
output.
Summary of the layers and parameters for the training of the model is shown in
Fig. 4.

4 Results and Analysis

The whole dataset is categorized into Hindi and English. The number of images of
the handwritings in the dataset was 108 writers handwriting of different category
based on their personality.
The general steps followed for all the experiments are:
5 Personality Prediction Through Handwriting … 65

Fig. 4 Summary of the architecture of the convolutional neural network

1. Data Augmentation
• Rescaling image
• Horizontal flip
• Shear range
• Zoom
2. Input Augmented Data
• Training data
• Testing data
3. Build the model
4. Compile the model
5. Fit the model
6. Evaluate the performance
66 G. Chaubey and S. K. Arjaria

4.1 Experiment 1

The Experiment 1 is done with the analysis of English handwritings. The English
handwriting of 110 writers containing the five big personality traits has the following
number of images in training (Table 1) and test (Table 2) dataset:
For training data:
For test data:
With the above data, the average accuracy obtained is 0.43184 and average loss
is 1.44 as shown below in Fig. 5.

Table 1 Number of training

Personality trait No. of handwritings
data for Experiment 1
Agreeableness 15
Conscientiousness 13
Extraversion 4
Neuroticism 18
Openness 38

Table 2 Number of testing

Personality trait No. of handwritings
data for Experiment 1
Agreeableness 4
Conscientiousness 4
Extraversion 1
Neuroticism 4
Openness 9

Fig. 5 Training and validation accuracy and loss graph of English handwriting
5 Personality Prediction Through Handwriting … 67

Table 3 Number of training

Personality trait No. of handwritings
data for Experiment 2
Agreeableness 15
Conscientiousness 14
Extraversion 4
Neuroticism 18
Openness 38

Table 4 Number of testing

Personality trait No. of handwritings
data for Experiment 2
Agreeableness 4
Conscientiousness 4
Extraversion 1
Neuroticism 4
Openness 9

4.2 Experiment 2

The Experiment 2 is done with the analysis of Hindi handwritings. The Hindi hand-
writing of 111 writers containing the five big personality traits has the following
number of images in training (Table 3) and test (Table 4) dataset:
For training data:
For test data:
With the above data, the average accuracy obtained is 42.63 and the average loss
is 1.44. The graph of training and validation accuracy is shown below in Fig. 6.

4.3 Experiment 3

The experiment 3 is done with analysis of mixture of all Hindi and English handwrit-
ings. The number of writers was 110 of English and 111 of Hindi writing. The mix
handwriting of 221 writers containing the five big personality traits has the following
number of images in training (Table 5) and test (Table 6) dataset:
For training data:
For test data:
With the above data, the average accuracy obtained is 42.94 and the average loss
is 1.42. The graph of training and validation accuracy is shown below in Fig. 7.
68 G. Chaubey and S. K. Arjaria

Fig. 6 Training and validation accuracy and loss graph of Hindi handwriting

Table 5 Number of training

Personality trait No. of handwritings
data for Experiment 3
Agreeableness 30
Conscientiousness 27
Extraversion 8
Neuroticism 36
Openness 76

Table 6 Number of testing

Personality trait No. of handwritings
data for Experiment 4
Agreeableness 8
Conscientiousness 8
Extraversion 2
Neuroticism 8
Openness 18

5 Conclusion and Future Scope

If the dataset is created on the basis of the labeling of the data by the graphologist,
then the accuracy obtained is more. The dataset can also be created by taking the
handwriting of the writers by creating the handwritings on creating a temporary
emotion by showing them a movie of that emotion. We can also train the model with
a large dataset to improve the model accuracy.
5 Personality Prediction Through Handwriting … 69

Fig. 7 Training and validation accuracy and loss graph English and Hindi (Mix) handwriting

References

1. H.N. Champa, K.R. Ananda kumar, Automated human behaviour prediction through hand-
writing analysis, in 2010 first International Conference on Integrated Intelligent Computing
(2010)
2. M. Gavrilescu, 3-layer Architecture for Determining the Personality Type from Handwriting
Analysis by Combining Neural Networks and Support Vector Machines, Department of
Telecommunication (2017)
3. G. Luria, S. Rosenblum, A Computerized Multidimensional Measurement of Mental Work Load
via Handwriting Analysis, Department of Human Services and Department of Occupational
Therepy (2012)
4. S. Mukherjee, I. De, Feature extraction from handwritten documents for personality analysis,
in 2016 International Conference on Computer, Electrical and Communication Engineering
(ICCECE) (2016)
5. M. Sachan, S.K. Singh, Personality detection using handwriting analysis, in Seventh Inter-
national Conference on Advance and Computing, Electronics and Communication (ACEC)
(2018)
6. B. Fallah, H. Khotanlou, Artificial intelligence and robotics (IRANOPEN), in Identify Human
Personality Parameters Based on Handwriting Using Neural Networks (April 2016)
7. H.N. Champa, K.R. Anandakumar, Automated human behavior prediction through handwriting
analysis, in 2010 First International Conference on Integrated Intelligent Computing (ICIIC),
pp. 160–165 (August 2010)
8. Z. Chen, T. Lin, Automatic personality identification using writing behaviors: an exploratory
study. Behav. Inform. Technol. 36(8), 839–845 (2017)
9. I. Siddiqi, C. Djeddi, A. Raza, L. Souici-Meslati, Automatic analysis of handwriting for gender
classification. Pattern. Anal. Appl. 18(4), 887–899 (2015)
10. S. Maadeed, A. Hassaine, Automatic prediction of age, gender, and nationality in offline
handwriting. EURASIP J. Image Video Process. 2014, 10 (2014)
11. G. Luria, S. Rosenblum, A computerized multidimensional measurement of mental workload
via handwriting analysis. Behav. Res. Methods 44(2), 575–586 (2012)
 70 G. Chaubey and S. K. Arjaria

12. I. Zaarour, L. Heutte, P. Leray, J. Labiche, B. Eter, D. Mellier, Clustering and Bayesian network
approaches for discovering handwriting strategies of primary school children. Int. J. Pattern
Recognit. Artif. Intell. 18(7), 1233–1251 (2004)
13. R. Sudirman, N. Tabatabaey-Mashadi, I. Ariffin, Aspects of a standardized automated system
for screening children’s handwriting, in First International Conference on Informatics and
Computational Intelligence (ICI), pp. 48–54 (December 2011)
14. G. Luria, A. Kahana, S. Rosenblum, Detection of deception via handwriting behaviors using a
computerized tool: toward an evaluation of malingering. Cogn. Comput. 6(4), 849–855 (2014)
15. S.B. Bhaskoro, S.H. Supangkat, An extraction of medical information based on human hand-
writings, in 2014 International Conference on Information Technology Systems and Innovation
(ICITSI), pp. 253–258 (November 2014)
16. P. Drotar, J. Mekyska, Z. Smekal, I. Rektorova, Prediction potential of different handwriting
tasks for diagnosis of Parkinson’s, in 2013 E-Health and Bioengineering Conference, pp. 1–4,
November 2013
17. W.L. Lee, K.-C. Fan, Document image preprocessing based on optimal Boolean filters. Signal
Process. 80(1), 45–55 (2000)
18. J.G. Leu, Edge sharpening through ramp width reduction. Image Vis. Comput. 18(6), 501–514
(2000)
19. S.C.F. Lin et al., Intensity and edge based adaptive unsharp masking filter for color image
enhancement. Optik Int. J. Light Electron Optics 127(1), 407–414 (2016)
20. R. Legault, C.Y. Suen, Optimal local weighted averaging methods in contour smoothing. IEEE
Trans. Pattern Anal. Mach. Intell. 18, 690–706 (1997)
21. Y. Solihin, C.G. Leedham, Integral ratio: a new class of global thresholding techniques for
handwriting images. IEEE Trans. Pattern Anal. Mach. Intell. 21, 761–768 (1999)
22. K. Chen, F. Yin, C.-L. Liu, Hybrid page segmentation with efficient whitespace rectangles
extraction and grouping, in 2013 12th International Conference on Document Analysis and
Recognition (ICDAR), pp. 958–962 (2013)
23. V. Papavassiliou, T. Stafylakis, V. Katsouro, G. Carayannis, Handwritten document image
segmentation into text lines and words. Pattern Recogn. 43, 369–377 (2010)
24. N. Lemos et al., Personality Prediction based on handwriting using machine learning, in 2018
International Conference on Computational Technique, Electronics and Mechanical Systems
(CTEMS), Dec. 2018
25. S.H. Fatimah et al., Personality features identification from handwriting using convolutional
neural network, in 4th International Conference on Information Technology, Information
Systems and Electrical Engineering (ICITISEE) (2019)
26. N. Nijil Raj, et al., Forecasting personality based on calligraphy using CNN and MLP. Int. J.
Comput. Sci. Eng. 8(7) (2020)

View publication stats