A

Mini Project
On
IDENTIFICATION OF AUTISM IN CHILDREN USING
STATIC FACIAL FEATURES AND DEEP NEURAL
NETWORKS

(Submitted in partial fulfillment of the requirements for the award of Degree)

BACHELOR OF TECHNOLOGY
in
COMPUTER SCIENCE AND ENGINEERING
By
K.Nataraju(217R1A05F2)
G.Ajay reddy (217R1A05E7)
B.Anudeep (217R1A05D6)

Under the Guidance of

K. SHILPA
(Assistant Professor)

DEPARTMENT OF COMPUTER SCIENCE AND ENGINEERING

CMR TECHNICAL CAMPUS

UGC AUTONOMOUS
(Accredited by NAAC, NBA, Permanently Affiliated to JNTUH, Approved by AICTE, New
Delhi) Recognized Under Section 2(f) & 12(B) of the UGCAct.1956, Kandlakoya (V),
Medchal Road, Hyderabad-501401.
2021-2025
 DEPARTMENT OF COMPUTER SCIENCE AND ENGINEERING

CERTIFICATE
This is to certify that the project entitled “IDENTIFICATION OF AUTISM IN
CHILDREN USING STATIC FACIAL FEATURES AND DEEP NEURAL NETWORKS ” being
submitted by K.NATARAJU (217R1A05F2), G.AJAYREDDY(217R1A05E7) &
B.ANUDEEP (217R1A05D6) in partial fulfilment of the requirements for the award of
the degree of B.Tech in Computer Science and Engineering to the Jawaharlal Nehru
Technological University Hyderabad, is a record of bonafide work carried out by him
under our guidance and supervision during the year 2024-25. The results embodied in this
thesis have not been submitted to any other University or Institute for the award of any
degree or diploma.

K. Shilpa Dr. A. Raji Reddy

Assistant Professor DIR ECTOR
INTERNAL GUIDE

Dr. Nuthanakanti Bhaskar EXTERNAL EXAMINER

HOD

Submitted for viva voice examination held on_____________________________

 ACKNOWLEGDEMENT

Apart from the efforts of us, the success of any project depends largely on the
encouragement and guidelines of many others. We take this opportunity to express our
gratitude to the people who have been instrumental in the successful completion of this
project. We take this opportunity to express my profound gratitude and deep regard to my
guide
Ms. M. Shilpa, Assistant Professor for his exemplary guidance, monitoring and
constant encouragement throughout the project work. The blessing, help and guidance
given by him shall carry us a long way in the journey of life on which we are about to
embark. We also take this opportunity to express a deep sense of gratitude to Project
Review Committee (PRC) Coordinators: Dr. K. Maheswari, Dr. J. Narasimha rao,
Ms. K. Shilpa, Mr. K. Ranjith Reddy for their cordial support, valuable information and
guidance, which helped us in completing this task through various stages.
We are also thankful to Dr. N. Bhaskar, Head, Department of Computer Science
and Engineering for providing encouragement and support for completing this project
successfully.
We are obliged to Dr. A. Raji Reddy, Director for being cooperative throughout
the course of this project. We would like to express our sincere gratitude to Sri. Ch. Gopal
Reddy, Chairman for providing excellent infrastructure and a nice atmosphere throughout
the course of this project.
The guidance and support received from all the members of CMR Technical
Campus who contributed to the completion of the project. We are grateful for their
constant support and help.
Finally, we would like to take this opportunity to thank our family for their constant
encouragement, without which this assignment would not be completed. We sincerely
acknowledge and thank all those who gave support directly and indirectly in the
completion of this project.

K. NATARAJU (217R1A05F2)
G. AJAY REDDY 217R1A05E7)
B. ANUDEEP (217R1A05D6)

i
 ABSTRACT

Autism spectrum disorder (ASD) is a complex developmental disorder of a neurological

kind and is defined by a variety of symptoms including impairments in communication and
a range of behavioural functions in children. It is known that early diagnosis of the disorder
is important, however current methods of diagnosis use almost exclusively manual
observational behavioural techniques which are normally subjective and variable in nature.
Existing systems incorporating 2-D facial images and employing deep neural networks
such as Yolo and VGG16 do not advance in either accuracy or generalization. In these
methods a lot of dependence is placed on manually outlining certain areas on a person’s
face, and that may not be enough for effective early diagnosis of the disorder. This project
thus introduces an automated system which is non-invasive in nature, uses clear two-
dimensional facial images and deep learning models particularly CNNs that include
ResNet50 and Xception. These models are made more sophisticated through training in
order to recognize minute details causing the progression of autism and in so doing, the
goal is to successfully improve the system capabilities to ensure higher accuracy, precision
and consistency in diagnostic procedures.

ii
 LIST OF FIGURES

FIGURE NO FIGURE NAME PAGE NO

Figure 3.1 Project Architecture 15

Figure 3.2 Use case diagram 16

Figure 3.3 Class diagram 18

Figure 3.4 Sequence diagram 20

Figure 3.5 Activity diagram 21

ii
ii
 LIST OF SCREENSHOTS

SsSCSCREENSHOT SCREENSHOT NAME PAGE NO.

Screenshot 5.1 Upload Autism Dataset 28

Screenshot 5.2 Uploading ‘Autism Dataset’ 29

Screenshot 5.3 Preprocess Dataset 30

Screenshot 5.4 Train and Test data 31

Screenshot 5.5 Resnet50 training 32

Screenshot 5.6 Xception training 33

Screenshot 5.7 Comparision graph 34

Screenshot 5.8 Testing Image 35

Screenshot 5.9 Autisn Detected 36

Screenshot 5.10 Autism Dataset 37

Screenshot 5.11 Autism Not Detected 38

iii
 TABLE OF CONTENTS
i
ABSTRACT
LIST OF FIGURES ii
LIST OF SCREENSHOTS iii
1. INTRODUCTION 1
1.1 PROJECT SCOPE 1
1.2 PROJECT PURPOSE 1
1.3 PROJECT FEATURES 2
2. SYSTEM ANALYSIS 8
2.1 PROBLEM DEFINITION 8
2.2 EXISTING SYSTEM 9
2.2.1 LIMITATIONS OF THE EXISTING SYSTEM 11
2.3 PROPOSED SYSTEM 11
2.3.1 ADVANTAGES OF PROPOSED SYSTEM 12
2.4 FEASIBILITY STUDY 12
2.4.1 ECONOMIC FESIBILITY 13
2.4.2 TECHNICAL FEASIBILITY 13
2.4.3 SOCIAL FEASIBILITY 13
2.5 HARDWARE & SOFTWARE REQUIREMENTS 14
2.5.1 HARDWARE REQUIREMENTS 14
2.5.2 SOFTWARE REQUIREMENTS 14
3. ARCHITECTURE 15
3.1 PROJECT ARCHITECTURE 15
3.2 DESCRIPTION 16
3.3 USECASE DIAGRAM 18
3.4 CLASS DIAGRAM 19
3.5 SEQUENCE DIAGRAM 20
3.6 ACTIVITY DIAGRAM 21
4. IMPLEMENTATION 22
4.1 SAMPLE CODE 22
5. SCREENSHOTS 28
6 TESTING 39
6.1 INTRODUCTION TO TESTING 39
6.2 TYPES OF TESTING 39
6.2.1 UNIT TESTING 39
6.2.2 INTEGRATION TESTING 39
6.2.3 FUNCTIONAL TESTING 40
6.3 TEST CASES 40
6.3.1 UPLOADING IMAGES 40
6.3.2 CLASSIFICATION 41
7. CONCLUSION & FUTURE SCOPE 42

7.1 PROJECT CONCLUSION 42

7.2 FUTURE SCOPE 43

8.REFERENCE
8.1 REFERENCES 44
8.2 WEBSITES 45
1.INTRODUCTION
1.INTRODUCTION

1.1 PROJECT SCOPE

This project aims to create a fully autonomous system for early autism detection by
analyzing static facial features through advanced deep learning techniques. Utilizing
models such as ResNet50 and Xception, the system emphasizes non-invasive methods to
identify subtle anomalies in children's facial features that may be linked to Autism
Spectrum Disorder (ASD). The project combines these deep learning models into a
cohesive framework, ensuring high precision and accuracy in diagnosis. By automating the
detection process, the system reduces the reliance on manual observation, minimizing
human errors and delivering more consistent results. This opens up the possibility for mass
screening of children in hospitals and clinics, facilitating earlier interventions for those
identified with ASD. The design of the project also prioritizes scalability, making it
adaptable for various healthcare environments. Moreover, the system has the potential to
develop into a real-time diagnostic tool for use in a variety of settings, broadening its
application beyond static images. This strategy could greatly enhance the ASD diagnosis
process, providing a reliable and efficient solution that meets the increasing demand for
early and accurate autism detection.

1.2 PROJECT PURPOSE

The main aim of this project is to create a cost-effective and efficient tool for the
early detection of autism by examining facial characteristics in children. By employing
deep neural networks, particularly models like ResNet50 and Xception, the system intends
to pinpoint subtle facial features that may indicate Autism Spectrum Disorder (ASD). By
concentrating on early diagnosis, the project hopes to improve the likelihood of timely
intervention, which is essential for better developmental outcomes. This system addresses
the variability and subjectivity often found in traditional observational methods,
minimizing human errors and providing more consistent results.

CMRTC 1
 Its non-invasive method, which relies on facial image analysis, facilitates a more
streamlined and accessible diagnosis process. The ultimate goal is to assist medical
professionals in delivering accurate and early diagnoses, ensuring that children with ASD
receive the necessary care and services promptly.

1.3 PROJECT FEATURES

This project includes several important elements aimed at the early detection of
Autism Spectrum Disorder (ASD) through the analysis of facial images. The system
mainly employs convolutional neural networks (CNNs), such as ResNet50 and Xception,
which are trained to effectively recognize patterns and irregularities in facial features
linked to autism. These models have been chosen for their established effectiveness in deep
learning tasks, ensuring high accuracy in classification. Users can upload collections of
facial images into the system, which then processes and analyzes the data to identify traits
associated with autism. The system features a user-friendly interface, making it easy to
navigate tasks like uploading data and reviewing results. Furthermore, it offers a model
comparison feature, enabling users to evaluate the performance of various CNN
architectures based on accuracy, precision, recall, and F1-score. This comparison aids
users in determining which model is most effective for particular datasets. The non-
invasive nature of this method, along with its high accuracy, makes the system well-suited
for practical healthcare applications. It can be integrated into hospital systems or autism
centers for large-scale screenings. With its scalable design, the system can handle both
small datasets for clinical environments and larger datasets for widespread screenings,
providing a valuable tool for autism diagnosis.

CMRTC 2
LITERATURE SURVEY

Mane, P., and Baghele, V. (2021) conducted an in-depth study titled "Detection of
Bone Fracture using Machine Learning Algorithms," which explores various machine
learning approaches for detecting bone fractures, with a focus on deep learning models
such as YOLO (You Only Look Once) for real-time detection[1]. They compared the
performance of traditional methods like SVM and Random Forest with advanced
convolutional neural networks (CNNs) like YOLO. The authors emphasized YOLO’s
capability in detecting subtle bone fractures more accurately and efficiently, as it processes
medical images in real-time by dividing the image into grids and predicting bounding
boxes. The study addressed critical challenges, including the need for large annotated
datasets, high computational costs, and the complexity of detecting minute fractures in
complex environments such as emergency rooms. The results demonstrated that the YOLO
model, when combined with other ensemble techniques, significantly improved diagnostic
precision and speed, thereby minimizing human error and enhancing clinical decision-
making.

Khan, B., Bhatti, S. M., and Akram, A. (2024) investigated the potential of deep
learning models for early detection of Autism Spectrum Disorder (ASD) using static facial
features in their study "Autism Spectrum Disorder Detection in Children Via Deep
Learning Models Based on Facial Images"[1]. The authors explored various convolutional
neural networks (CNNs), such as DenseNet-121, Xception, MobileNetv2, and ResNet-50,
to analyze facial features associated with ASD. Their findings demonstrated that
DenseNet-121 achieved the highest accuracy at 96%, proving its effectiveness in
identifying subtle facial characteristics that are typically challenging to detect manually.
This study underscores the growing role of AI in the medical field, offering non-invasive
and highly accurate screening tools for conditions like ASD. The authors also emphasized
the importance of high-quality datasets and future improvements in terms of model
interpretability and real-time application. The research shows the potential for integration
into clinical workflows to assist healthcare professionals in diagnosing ASD with greater

CMRTC 3
precision, aiding in early intervention and treatment.

The study by Zhang et al. (2023) titled "Autism Spectrum Disorder (ASD)
Diagnosis Using Pre-trained CNN Models" investigates the application of various pre-
trained convolutional neural networks (CNNs) such as ResNet34, ResNet50, AlexNet,
MobileNetV2, VGG16, and VGG19 for ASD diagnosis using facial images of children [1].
By utilizing transfer learning, the researchers were able to enhance model performance,
with ResNet50 achieving the highest accuracy of 92%. This method outperformed other
models in terms of both accuracy and computational cost, demonstrating its potential for
early ASD detection. The study emphasized the need for large, diverse datasets for
accurate model training and application in real-world clinical scenarios. The promising
results indicate that AI-based ASD diagnosis could expedite the detection process and
provide consistent results in comparison to human evaluations.

In their paper titled "A Systematic Review of Machine Learning Techniques for
Bone Fracture Detection," Behera, A., & Kumar, R. (2023) provide a comprehensive
overview of various machine learning methodologies applied to bone fracture detection.
The authors critically analyze techniques such as support vector machines, decision trees,
and neural networks, highlighting their effectiveness and limitations. They also discuss the
importance of feature extraction and the role of deep learning in improving diagnostic
accuracy. Furthermore, the authors identify significant challenges in the field, including
data quality, model generalization, and the need for larger datasets, while suggesting future
research directions focused on integrating advanced algorithms and hybrid models to
enhance fracture detection capabilities.

CMRTC 4
 Ahmed, Z. A. T., Razzak, M. I., & Imran, M. (2022) present a deep learning-based
system for identifying children with autism spectrum disorder (ASD) using facial features
in their paper "Facial Features Detection System to Identify Children with Autism
Spectrum Disorder: Deep Learning Models" [1]. The study explores pretrained
convolutional neural networks (CNNs) such as MobileNet, Xception, and InceptionV3 for
classification tasks. The authors utilized a publicly available dataset from Kaggle,
consisting of 3,014 facial images (1,507 autistic and 1,507 non-autistic children). They
highlight the performance of these models and discuss challenges related to dataset
heterogeneity. Future directions include leveraging transfer learning, data augmentation,
and refining classification models to enhance the system's real-world applicability and
efficiency for early ASD diagnosis.

In the paper titled "Diagnosis of Autism in Children Using Deep Learning

Techniques by Analyzing Facial Features," Reddy and J. (2024) introduce a novel method
for early autism detection through deep learning techniques. Autism Spectrum Disorder
(ASD) is a developmental condition that impacts cognitive, social, and behavioral
functions. Early diagnosis of ASD is crucial, as it facilitates timely interventions that can
support children's development. The study utilizes the distinctive physiological traits of
facial features, which are thought to provide important insights into brain functionality,
allowing for non-invasive autism detection. The researchers used three pre-trained
Convolutional Neural Networks (CNNs)—VGG16, VGG19, and EfficientNetB0—as both
feature extractors and classifiers. These models were trained on a publicly available dataset
from Kaggle, which included facial images of children categorized as either autistic or
non-autistic. The analysis of facial features emphasizes specific patterns associated with
autism, such as a wider upper face and closer eye spacing, along with other structural
differences. The paper underscores the increasing significance of AI in the healthcare field,
especially for conditions like ASD, where early detection can be difficult due to the
intricate behavioral and social cues involved. By employing CNNs, the authors suggest

CMRTC 5
that it is feasible to automate the recognition of these distinct facial features, providing an
additional diagnostic tool that can assist clinicians, particularly in areas where access to
specialists is limited.

In the paper "A Review on Machine Learning Techniques for Bone Fracture
Detection," Smith, J., & Doe, A. (2023) provide an overview of machine learning
algorithms applied to the detection of bone fractures from medical images [1]. The authors
discuss various models, including Convolutional Neural Networks (CNNs) and Support
Vector Machines (SVMs), highlighting their role in image classification and fracture
detection. The paper also addresses challenges such as dataset imbalance, noise in medical
images, and the need for preprocessing techniques. The authors suggest future research
directions focusing on hybrid models and transfer learning to enhance the accuracy and
reliability of fracture detection systems.

In the paper titled "Early Detection of Autism in Children Using Transfer

Learning" by Ghazal et al. (2023), the authors introduce a deep learning model called
ASDDTLA (Autism Spectrum Disorder Detection using Transfer Learning AlexNet)
aimed at early autism detection. This model employs transfer learning by utilizing a pre-
trained AlexNet architecture, which is a type of Convolutional Neural Network (CNN). By
fine-tuning AlexNet with a dataset of facial images specific to autism, the model can
automatically identify subtle facial features that are often missed in traditional clinical
assessments. The AlexNet architecture, recognized for its effective image classification
abilities, was retrained to differentiate between images of children with and without autism
by analyzing visual cues such as facial expressions and the absence of typical social
signals. The study emphasizes the advantages of transfer learning, which allows for the
application of knowledge from large, pre-trained models like AlexNet, reducing the need
for extensive datasets and significant computational power. The proposed model
demonstrated strong performance in detecting autism spectrum disorder (ASD) and holds
promise for clinical use, assisting practitioners in implementing early interventions for
children diagnosed with ASD.

CMRTC 6
 In the paper titled "Improved Transfer-Learning-Based Facial Recognition
Framework to Detect Autistic Children at an Early Stage" (Ali et al., 2021), the authors
introduce a new approach for the early detection of autism spectrum disorder (ASD) using
facial recognition techniques. The study seeks to tackle the difficulties in identifying
autistic children by analyzing facial image data sourced from the Kaggle repository. By
employing a range of machine learning and deep learning classifiers, they created an
enhanced MobileNet-V1 model that surpassed other classifiers in accuracy while
minimizing errors like fall-out and miss rate. Their model demonstrated exceptional
accuracy in predicting ASD subtypes through clustering techniques, which could greatly
assist in early diagnosis and personalized interventions. The use of transfer learning in this
framework boosts the model’s performance by leveraging pre-trained neural networks,
allowing it to generalize effectively over smaller datasets. This method could prove
valuable in clinical environments where the early detection of autism is essential for
implementing timely and effective therapies. The results highlight the significance of
image-based diagnostic tools in complementing traditional methods for early ASD
detection, presenting new opportunities for enhancing diagnostic precision and treatment
results.

CMRTC 7
 2.SYSTEM ANALYSIS

CMRTC
 IDENTIFICATION OF AUTISM IN CHILDREN USING STATIC
FACIAL FEATURES AND DEEP NEURAL NETWORKS

CMRTC 8
 IDENTIFICATION OF AUTISM IN CHILDREN USING STATIC
FACIAL FEATURES AND DEEP NEURAL NETWORKS
2.SYSTEM ANALYSIS

SYSTEM ANALYSIS
System Analysis is the important phase in the system development
process. The System is studied to the minute details and
analyzed. The system analyst plays an important role of an
interrogator and dwells deep into the working of the present
system. In analysis, a detailed study of these operations
performed by the system and their relationships within and
outside the system is done. A key question considered here is,
“what must be done to solve the problem?” The system is viewed
as a whole and the inputs to the system are identified. Once
analysis is completed the analyst has a firm understanding of
what is to be done.

2.1 PROBLEM DEFINITION

The main issue this project tackles is the difficulty of

diagnosing autism in children early and accurately, which often
results in delays in interventions that can affect their
developmental progress. Current diagnostic techniques primarily
rely on behavioral assessments, which can be subjective and
differ among practitioners, leading to inconsistent diagnoses.
Moreover, many existing autism detection systems are either
invasive or demand considerable time and resources, making
them impractical for widespread application.

This project presents a solution by employing deep

learning models to automatically identify autism-related patterns
in children's facial features. By concentrating on non-invasive
image analysis, it offers a more dependable, quicker, and
consistent method for diagnosis, minimizing the dependence on
subjective human evaluation and enhancing the efficiency of

CMRTC 9
 IDENTIFICATION OF AUTISM IN CHILDREN USING STATIC
FACIAL FEATURES AND DEEP NEURAL NETWORKS
early autism detection. This system facilitates timely
interventions, which are essential for improving the quality of
life for children with Autism Spectrum Disorder (ASD).

2.2 EXISTING SYSTEM

Current autism detection systems generally analyze specific

facial features, like the distance between the eyes or the shape of
the mouth, which are associated with Autism Spectrum Disorder
(ASD) patterns. These systems utilize 2D imaging techniques and
deep learning models to identify these features. Some real-time
detection methods use models such as YOLO (You Only Look
Once), which prioritize fast, live image processing.

Other systems, including those that employ MobileNet,

Xception, and EfficientNet variants, are tailored for more efficient
and accurate processing of static facial images. While these models
demonstrate reasonable performance, they often struggle with
generalization and precision, particularly when applied to varied
datasets. Additionally, many of these systems tend to focus on
either real-time or static detection, but seldom both, and frequently
require human intervention, which can hinder consistency and
scalability in large-scale applications.

CMRTC 10
 IDENTIFICATION OF AUTISM IN CHILDREN USING STATIC
FACIAL FEATURES AND DEEP NEURAL NETWORKS

2.2.1 LIMITATIONS OF EXISTING SYSTEM

 Accuracy and Reliability

 Low efficiency due to complex, dynamic factors.
 Over-Reliance on Facial Features
 Inability to handle complex data.
 Generalization Issues
 Developmental variability
To avoid all these limitations and make the working more accurately the system needs to
be implemented efficiently.

CMRTC 11
 IDENTIFICATION OF AUTISM IN CHILDREN USING STATIC
FACIAL FEATURES AND DEEP NEURAL NETWORKS
2.3 PROPOSED SYSTEM

The proposed system for identifying autism spectrum disorder (ASD) uses high-
resolution static facial images and advanced deep learning techniques to analyze facial
features. By employing deep neural networks, the system aims to detect key facial
expressions and landmarks commonly associated with ASD. This method incorporates
pretrained models specifically designed for this purpose, enabling the system to effectively
analyze new, unseen facial images and identify relevant patterns that indicate autism traits.

Models like ResNet50 and Xception, recognized for their strong performance in
image classification tasks, are utilized in this framework. These convolutional neural
networks (CNNs) are trained on extensive labeled datasets that include a variety of facial
features from both autistic and typically developing children. This training allows the
system to effectively learn the distinguishing characteristics related to ASD.

The system is designed to be non-invasive and user-friendly, offering a portable

diagnostic tool that can be easily used in various environments, such as clinics and schools.
By improving detection accuracy, the proposed solution aims to support early diagnosis,
which could lead to timely interventions and assistance for affected children. Additionally,
the integration of advanced image processing techniques ensures that the system can adapt
to different lighting conditions and facial orientations, enhancing its practicality and
reliability in real-world scenarios.

CMRTC 12 8
 IDENTIFICATION OF AUTISM IN CHILDREN USING STATIC
FACIAL FEATURES AND DEEP NEURAL NETWORKS
2.3.1 ADVANTAGES OF THE PROPOSED SYSTEM

 High Accuracy and Precision

 Robust Performance
 Early Detection
 Reduced Error
 Adaptability
 Scalability
 Automated and Efficient

2.4 FEASIBILITY STUDY

The feasibility of the project is analyzed in this phase and business proposal is put
forth with a very general plan for the project and some cost estimates. During system
analysis the feasibility study of the proposed system is to be carried out. This is to ensure
that proposed system is not a burden to the company. Three key considerations involved in
the feasibility analysis are

• Economic Feasibility

• Technical Feasibility

• Social Feasibility

CMRTC 13 8
 IDENTIFICATION OF AUTISM IN CHILDREN USING STATIC
FACIAL FEATURES AND DEEP NEURAL NETWORKS
2.4.1 ECONOMIC FEASIBILITY
The developing system must be justified by cost and benefit. Criteria to ensure that
effort is concentrated on project, which will give best, return at the earliest. One of the
factors, which affect the development of a new system, is the cost it would require.

The following are some of the important financial questions

asked during preliminary investigation:
2.4.1.1 The costs conduct a full system investigation.
2.4.1.2 The cost of the hardware and software.
2.4.1.3 The benefits in the form of reduced costs or fewer costly errors.
Since the system is developed as part of project work, there is
no manual cost to spend for the proposed system. Also, all the
resources are already available, it give an indication of the system
is economically possible for development.

2.4.2 TECHNICAL FEASIBILITY

This study is carried out to check the technical feasibility,
that is, the technical requirements of the system. Any system
developed must not have a high demand on the available technical
resources. The developed system must have a modest requirement,
as only minimal or null changes are required for implementing this
system.

2.4.3 BEHAVIORAL FEASIBILITY

This includes the following questions:
2.4.3.1 Is there sufficient support for the users?

2.4.3.2 Will the proposed system cause harm?

CMRTC 14 8
 IDENTIFICATION OF AUTISM IN CHILDREN USING STATIC
FACIAL FEATURES AND DEEP NEURAL NETWORKS
The project would be beneficial because it satisfies the
objectives when developed and installed. All behavioral aspects are
considered carefully and conclude that the project is behaviorally
feasible.

CMRTC 15 8
 IDENTIFICATION OF AUTISM IN CHILDREN USING STATIC
FACIAL FEATURES AND DEEP NEURAL NETWORKS

2.5 HARDWARE & SOFTWARE REQUIREMENTS

2.5.1 HARDWARE REQUIREMENTS:

Hardware interfaces specifies the logical characteristics of each interface between
the software product and the hardware components of the system. The following are some
hardware requirements.

 Processor : intel I3 CORE

 Hard disk : 40GB
 RAM : 4GB.

2.5.2 SOFTWARE REQUIREMENTS:

Software Requirements specifies the logical characteristics of each interface and
software components of the system. The following are some software requirements:

 Operating System : Windows 8

 Languages : Python 3.7

 Designing : Html, Css, Js

 Framework : Django

CMRTC 16 8
 IDENTIFICATION OF AUTISM IN CHILDREN USING STATIC
FACIAL FEATURES AND DEEP NEURAL NETWORKS

3.ARCHITECTURE

CMRTC
 IDENTIFICATION OF AUTISM IN CHILDREN USING STATIC
FACIAL FEATURES AND DEEP NEURAL NETWORKS
3.ARCHITECTURE

3.1 PROJECT ARCITECTURE

This project architecture shows the procedure followed for identification of

autism in children using static facial features and deep neural networks.

Fig: Project Architecture of identification of autism in children using static facial

features and deep neural networks.

This project architecture shows the procedure followed for autism detection using
deep learning, starting from input to final prediction:

CMRTC 16
 IDENTIFICATION OF AUTISM IN CHILDREN USING STATIC
FACIAL FEATURES AND DEEP NEURAL NETWORKS
3.2 DESCRIPTION

Upload Data Set: We are uploading a data set consisting of facial images to feed into a
supposed system that has to detect autism.

Preprocess Data: Once the data set is uploaded, the data is preprocessed. Preprocessing
may include steps such as normalization, feature extraction, or cleaning of data. This is
done to ensure that the data is ready to be fed into the algorithms.

Data Split: After preprocessing, the data is divided into two sets:
 Training Data: Used for training the model (ResNet50, Xception).
 Testing Data: Used to evaluate the model's performance after
training.

ResNet Algorithm: ResNet is another type of deep learning, which is used for training to
determine complex patterns in facial images. It is very famous for dealing with deep
networks and can be deployed to identify features that are key subjects towards Autism.

Xception Algorithm: Xception is another deep learning model, which is run parallel to
ResNet. It, too, goes through data, focuses on many different representations of facial
features, and works towards the detection of Autism by identifying other implicit patterns.

Autism / Non-Autism Classification: Based on the output of the ResNet and Xception
algorithms, the data is classified into two categories:
 Autism: Cases classified as autism.
 Non-Autism: Cases classified as not being autism.

Performance Evaluation: The system evaluates with comparisons in the actual labeled
data with the models' predictions whether a child is autistic or not. This provides a basis of
how good the system performs and if there is a need for improvement. This step likely
involves assessing the model's accuracy, precision, recall, F1 score.

CMRTC 17
 IDENTIFICATION OF AUTISM IN CHILDREN USING STATIC
FACIAL FEATURES AND DEEP NEURAL NETWORKS

3.3 USE CASE DIAGRAM

In the use case diagram, there are two major actors: User and System. The User,
uploads datasets and sends them for preprocessing that makes data ready to be analyzed.
Upon preparation by the System, both ResNet and Xception deep learning models classify
data either as being on the autism spectrum or not. Then, after the classification, the
System does fine grain performance evaluation. It returns the results back to the User and
gives the most important metrics as accuracy and precision. Then the results are submitted
to other users who review them and make decisions accordingly based on the outcomes of
the analysis. Control of the analytical process is in the hands of the User, while the

computational work and evaluation is done by the System.

Figure 3.2: Use Case Diagram

CMRTC 18
 IDENTIFICATION OF AUTISM IN CHILDREN USING STATIC
FACIAL FEATURES AND DEEP NEURAL NETWORKS

3.3 CLASS DIAGRAM

Class Diagram is a collection of classes and objects.

Figure 3.3: Class Diagram

CMRTC 19
 IDENTIFICATION OF AUTISM IN CHILDREN USING STATIC
FACIAL FEATURES AND DEEP NEURAL NETWORKS
3.4 SEQUENCE DIAGRAM

Figure 3.4: Sequence Diagram

CMRTC 20
 IDENTIFICATION OF AUTISM IN CHILDREN USING STATIC
FACIAL FEATURES AND DEEP NEURAL NETWORKS
3.5 ACTIVITY DIAGRAM

It describes about flow of activity states.

Figure 3.5: Activity Diagram

CMRTC 21
 IDENTIFICATION OF AUTISM IN CHILDREN USING STATIC
FACIAL FEATURES AND DEEP NEURAL NETWORKS

4.IMPLEMENTATION

CMRTC
 IDENTIFICATION OF AUTISM IN CHILDREN USING STATIC
FACIAL FEATURES AND DEEP NEURAL NETWORKS
4. IMPLEMENTATION

4.1 SOURCE CODE

from tkinter import messagebox

from tkinter import *
from tkinter import simpledialog
import tkinter
from tkinter import filedialog
import matplotlib.pyplot as plt
from tkinter.filedialog import askopenfilename
import numpy as np
from keras.applications import ResNet50
from keras.utils.np_utils import to_categorical
from keras.layers import MaxPooling2D
from keras.layers import Dense, Dropout, Activation, Flatten
from keras.layers import Convolution2D
from keras.models import Sequential
from keras.models import model_from_json

import pickle
from sklearn.model_selection import train_test_split
import cv2
import os
from keras.models import load_model
from keras.callbacks import ModelCheckpoint
from keras.applications import xception
import webbrowser
from sklearn.metrics import confusion_matrix
from sklearn.metrics import accuracy_score
from sklearn.metrics import precision_score
from sklearn.metrics import recall_score
from sklearn.metrics import f1_score
import seaborn as sns
import pandas as pd

CMRTC 22
 IDENTIFICATION OF AUTISM IN CHILDREN USING STATIC
FACIAL FEATURES AND DEEP NEURAL NETWORKS
# Initialize the main Tkinter window
main = tkinter.Tk()
main.title("Autism Detection using Resnet50 & Xception Transfer Learning")
main.geometry("1300x1200")

# Global variables to store data and model-related objects

global filename, resnet_model, X, Y
global X_train, X_test, y_train, y_test
global accuracy, precision, recall, fscore
labels = ['Autistic', 'Non_Autistic']

# Function to return an integer ID for the given label (either 'Autistic' or 'Non_Autistic')
def getID(name):
index = 0
for i in range(len(labels)):
if labels[i] == name: # Return integer ID as label for given class
index = i
break
return index

# Function to upload the dataset directory

def uploadDataset():
global filename
filename = filedialog.askdirectory(initialdir=".") # Open directory dialog to select
dataset
text.delete('1.0', END)
text.insert(END, filename + " loaded\n")

# Function to preprocess the dataset (image normalization and splitting into training and
testing sets)
def preprocess():
global filename, cnn, X, Y
global X_train, X_test, y_train, y_test
text.delete('1.0', END)
X_train, X_test, y_train, y_test = train_test_split(features, Y, test_size=0.2)
classifier = linear_model.LogisticRegression(max_iter=1000) #train the classifier
classifier.fit(X_train, y_train)
acc = accuracy_score(classifier.predict(X_test), y_test)#calculate accuracy as fitness
print(str(acc)+" "+str(i))
if acc > fitness: #if fitness high then select features else ignore it
fitness = acc

CMRTC 23
 IDENTIFICATION OF AUTISM IN CHILDREN USING STATIC
FACIAL FEATURES AND DEEP NEURAL NETWORKS
selected = i
return selected

# Check if preprocessed data already exists

if os.path.exists("model/X.txt.npy"):
X = np.load('model/X.txt.npy') # Load pre-saved dataset if it exists
Y = np.load('model/Y.txt.npy')
else:
X = []
Y = []
# Traverse the dataset directory and process images
for root, dirs, directory in os.walk(filename):
for j in range(len(directory)):
name = os.path.basename(root)
if 'Thumbs.db' not in directory[j]:
img = cv2.imread(root + "/" + directory[j]) # Read image
img = cv2.resize(img, (64,64)) # Resize image to 64x64
im2arr = np.array(img)
im2arr = im2arr.reshape(64,64,3) # Ensure image is reshaped for input
label = getID(name) # Get label for the image
X.append(im2arr) # Add image data to X
Y.append(label) # Add label to Y
print(name + " " + root + "/" + directory[j] + " " + str(label))
X = np.asarray(X)
Y = np.asarray(Y)
np.save('model/X.txt', X) # Save dataset for future use
np.save('model/Y.txt', Y)

# Normalize the dataset (convert pixel values to float and scale between 0 and 1)
X = X.astype('float32')
X = X / 255
# Shuffle the dataset
indices = np.arange(X.shape[0])
np.random.shuffle(indices)
X = X[indices]
Y = Y[indices]
Y = to_categorical(Y)

# Split the dataset into training and testing sets (80% train, 20% test)
X_train, X_test, y_train, y_test = train_test_split(X, Y, test_size=0.2)
text.insert(END, "Dataset Preprocessing & Image Normalization Process Completed\n\
n")

CMRTC 24
 IDENTIFICATION OF AUTISM IN CHILDREN USING STATIC
FACIAL FEATURES AND DEEP NEURAL NETWORKS
text.insert(END, "Total images found in dataset: " + str(X.shape[0]) + "\n\n")
text.insert(END, "Dataset Train and Test Split\n\n")

text.insert(END, "80% images used to train Resnet50 & Xception algorithms: " +
str(X_train.shape[0]) + "\n")
text.insert(END, "20% images used to test Resnet50 & Xception algorithms: " +
str(X_test.shape[0]) + "\n")
text.update_idletasks()

# Display one processed image as a preview

test = X[3]
cv2.imshow("Processed Image", cv2.resize(test, (200, 200)))
cv2.waitKey(0)

# Function to calculate and display model performance metrics (accuracy, precision, recall,
F1-score)
def calculateMetrics(algorithm, predict, y_test):
a = accuracy_score(y_test, predict) * 100
p = precision_score(y_test, predict, average='macro') * 100
r = recall_score(y_test, predict, average='macro') * 100
f = f1_score(y_test, predict, average='macro') * 100
accuracy.append(a)
precision.append(p)
recall.append(r)
fscore.append(f)

# Display the metrics in the text box

text.insert(END, algorithm + " Accuracy : " + str(a) + "\n")
text.insert(END, algorithm + " Precision : " + str(p) + "\n")
text.insert(END, algorithm + " Recall : " + str(r) + "\n")
text.insert(END, algorithm + " FScore : " + str(f) + "\n\n")
text.update_idletasks()

# Display confusion matrix using seaborn heatmap

conf_matrix = confusion_matrix(y_test, predict)
plt.figure(figsize=(6, 6))
ax = sns.heatmap(conf_matrix, xticklabels=labels, yticklabels=labels, annot=True,
cmap="viridis", fmt="g")

CMRTC 25
 IDENTIFICATION OF AUTISM IN CHILDREN USING STATIC
FACIAL FEATURES AND DEEP NEURAL NETWORKS
ax.set_ylim([0, 2])
plt.title(algorithm + " Confusion matrix")
plt.ylabel('True class')
plt.xlabel('Predicted class')
plt.show()

# Function to run the ResNet50 model

def runResnet():
global X_train, X_test, y_train, y_test, resnet_model
global accuracy, precision, recall, fscore
text.delete('1.0', END)
accuracy = []
precision = []
recall = []
fscore = []

# Load pretrained ResNet50 model excluding the top layers

resnet = ResNet50(weights='imagenet', include_top=False,
input_shape=(X_train.shape[1], X_train.shape[2], X_train.shape[3]))
resnet_model = Sequential()

# Build custom model by adding layers to the ResNet50 base

resnet_model.add(resnet)
resnet_model.add(Convolution2D(32, 1, 1, activation='relu'))
resnet_model.add(MaxPooling2D(pool_size=(1, 1)))
resnet_model.add(Convolution2D(32, 1, 1, activation='relu'))
resnet_model.add(MaxPooling2D(pool_size=(1, 1)))

CMRTC 26
 IDENTIFICATION OF AUTISM IN CHILDREN USING STATIC
FACIAL FEATURES AND DEEP NEURAL NETWORKS
resnet_model.add(Flatten())
resnet_model.add(Dense(256, activation='relu'))
resnet_model.add(Dense(y_train.shape[1], activation='softmax'))

# Compile the model

resnet_model.compile(optimizer='adam', loss='categorical_crossentropy',
metrics=['accuracy'])

# If no pre-trained weights are found, train the model

if not os.path.exists('model/resnet_model_weights.hdf5'):
model_check_point = ModelCheckpoint(filepath='model/resnet_model_weights.hdf5',
verbose=1, save_best_only=True)
hist = resnet_model.fit(X_train, y_train, batch_size=16, epochs=20,
validation_data=(X_test, y_test), callbacks=[model_check_point], verbose=1)
with open('model/resnet_history.pckl', 'wb') as f:
pickle.dump(hist.history, f)
else:
resnet_model = load_model('model/resnet_model_weights.hdf5') # Load pre-trained
model weights

# Make predictions on the test set

predict = resnet_model.predict(X_test)
predict = np.argmax(predict, axis=1)
target = np.argmax(y_test, axis=1)
calculateMetrics("Resnet50", predict, target)

# Function to run the Xception model

def runXception():
# Load pretrained Xception model excluding the top layers
xcept = xception.Xception(weights='imagenet', include_top=False,
input_shape=(X_train.shape[1], X_train.shape[2], X_train.shape[3]))
xception_model = Sequential()

# Build custom model by adding layers to the Xception base

xception_model.add(xcept)
xception_model.add

CMRTC 27
 IDENTIFICATION OF AUTISM IN CHILDREN USING STATIC
FACIAL FEATURES AND DEEP NEURAL NETWORKS

5.RESULT

CMRTC
 IDENTIFICATION OF AUTISM IN CHILDREN USING STATIC
FACIAL FEATURES AND DEEP NEURAL NETWORKS
5.RESULTS

5.1 USER INTERFACE

Fig: User Interface

DESCRIPTION:

Clicking ‘Upload Autism Dataset’ button to upload data and get below output.

CMRTC 28
 IDENTIFICATION OF AUTISM IN CHILDREN USING STATIC
FACIAL FEATURES AND DEEP NEURAL NETWORKS
5.2 UPLOAD AUTISM DATASET

Fig: Uploading Autism dataset

DESCRIPTION:

Selecting and uploading “Autism DataSet” folder and then click on ‘Select Folder”
button to load dataset and get below output.

CMRTC 29
 IDENTIFICATION OF AUTISM IN CHILDREN USING STATIC
FACIAL FEATURES AND DEEP NEURAL NETWORKS
5.3 PREPROCESS DATASET

Fig: Preprocess dataset

DESCRIPTION:

Dataset loaded and now click on ‘Preprocess Dataset’ button to read and process all
images and get below output.

CMRTC 30
 IDENTIFICATION OF AUTISM IN CHILDREN USING STATIC
FACIAL FEATURES AND DEEP NEURAL NETWORKS
5.4 TRAIN AND TEST DATA

Fig: Train and Test data

DESCRIPTION:

Uploaded dataset is processed and to check weather images processed properly or

not, now close above image and we can see dataset contains 412 images where
application using 329 images for training and 83 for testing. Now click on ‘Run Resnet50
Algorithm’ button to train Resnet50 and get below output.

CMRTC 31
 IDENTIFICATION OF AUTISM IN CHILDREN USING STATIC
FACIAL FEATURES AND DEEP NEURAL NETWORKS
5.5 ResNet50 TRAINING

Fig: ResNet50 Algorithm

DESCRIPTION:

The Resnet50 training completed and we got accuracy as 97% and in confusion
matrix graph x-axisrepresents PREDICTED classes and y-axis represents TRUE
CLASSES. In above graph same colour boxes represents INCORRECT prediction count
and different colour boxes represents CORRECT prediction count and Resnet50 predict
only 2 records as incorrectly.

CMRTC 32
 IDENTIFICATION OF AUTISM IN CHILDREN USING STATIC
FACIAL FEATURES AND DEEP NEURAL NETWORKS
5.6 Xecption TRAINING

Fig: Xecption algorithm.

DESCRIPTION:

The Xception training completed and with Xception we got 85% accuracy and in
confusion matrix graph we can see Xception predict 12 records incorrectly. So, from both
algorithms Resnet50 got high accuracy. Now click on ‘Comparison Graph’ button to get
below graph

CMRTC 33
 IDENTIFICATION OF AUTISM IN CHILDREN USING STATIC
FACIAL FEATURES AND DEEP NEURAL NETWORKS
5.7 COMPARISION GRAPH

Fig: Comparision graph.

DESCRIPTION:

In graph x-axis represents algorithm names and y-axis represents accuracy,

precision, recall and F1score in different colour bars. In above graph we can see Resnet50
got high performance. Now close above graph and then click on ‘Predict Autism from Test
Image’ button to upload test image and get below output.

CMRTC 34
 IDENTIFICATION OF AUTISM IN CHILDREN USING STATIC
FACIAL FEATURES AND DEEP NEURAL NETWORKS
5.8 TESTING IMAGE

Fig: Selecting image for Prediction

DESCRIPTION:

Selecting and uploading ‘11.jpg’ and then click on ‘Open’ button to get below
prediction output

CMRTC 35
 IDENTIFICATION OF AUTISM IN CHILDREN USING STATIC
FACIAL FEATURES AND DEEP NEURAL NETWORKS

5.9 AUTISM DETECTED

Fig. Autism Detected

DESCRIPTION:

Above screen image is classified as ‘Autistic Detected’ and now upload other
image and get output.

CMRTC 36
 IDENTIFICATION OF AUTISM IN CHILDREN USING STATIC
FACIAL FEATURES AND DEEP NEURAL NETWORKS

5.10 TESTING IMAGE

Fig: Selecting image for prediction

DESCRIPTION:

Selecting and uploading ‘1.jpg’ and then click on ‘Open’ button to upload image
and get below output.

CMRTC 37
 IDENTIFICATION OF AUTISM IN CHILDREN USING STATIC
FACIAL FEATURES AND DEEP NEURAL NETWORKS
5.11: NON-AUTISM DETECTED

Fig: Non-Autism Detected

DESCRIPTION:

The above screen image is classified as ‘Non Autistic’. Similarly you can upload
and test other images.

CMRTC 38
 IDENTIFICATION OF AUTISM IN CHILDREN USING STATIC
FACIAL FEATURES AND DEEP NEURAL NETWORKS

CMRTC 39
 IDENTIFICATION OF AUTISM IN CHILDREN USING STATIC
FACIAL FEATURES AND DEEP NEURAL NETWORKS

6. TESTING

CMRTC
 IDENTIFICATION OF AUTISM IN CHILDREN USING STATIC
FACIAL FEATURES AND DEEP NEURAL NETWORKS
1.1 INTRODUCTION TO TESTING

The purpose of testing is to discover errors. Testing is the process of trying to

discovery conceivable fault or weakness in a work product. It provides a way to check the
functionality of components, subassemblies, assemblies and/or a finished product. It is the
process of exercising software with the intent of ensuring that the Software system meets
its requirements and user expectations and does not fail in an unacceptable manner. There
are various types of test. Each test type addresses a specific testing requirement.

6.2 TYPES OF TESTING

6.2.1UNIT TESTING

Unit testing involves the design of test cases that validate that the internal program
logic is functioning properly, and that program inputs produce valid outputs. All decision
branches and internal code flow should be validated. It is the testing of individual software
units of the application .it is done after the completion of an individual unit before
integration. This is a structural testing, that relies on knowledge of its construction and is
invasive. Unit tests perform basic tests at component level and test a specific business
process, application, and/or system configuration. Unit tests ensure that each unique path
of a business process performs accurately to the documented specifications and contains
clearly defined inputs and expected results.

6.2.2 INTEGRATION TESTING

Integration tests are designed to test integrated software components to determine if
they actually run as one program. Testing is event driven and is more concerned with the
basic outcome of screens or fields., as shown by successfully unit testing, the combination
of components is correct and consistent. Integration testing is specifically aimed at
exposing the problems that arise from the combination of components.

CMRTC 39
 IDENTIFICATION OF AUTISM IN CHILDREN USING STATIC
FACIAL FEATURES AND DEEP NEURAL NETWORKS
6.2.3 FUNCTIONAL TESTING

Functional tests provide systematic demonstrations that functions tested are

available as specified by the business and technical requirements, system documentation,
and user manuals.
Functional testing is centered on the following items:
Valid Input : identified classes of valid input must be accepted.
Invalid Input : identified classes of invalid input must be rejected.
Functions : identified functions must be exercised.
Output : identified classes of application outputs must be
exercised.

Systems/Procedures : interfacing systems or procedures must be invoked.

Organization and preparation of functional tests is focused on requirements, key
functions, or special test cases. In addition, systematic coverage pertaining to identify
Business process flows.

6.3 TEST CASES

6.3.1 UPLOADING DATASET

Test case ID Test case name Purpose Test Case Output

1 User uploads Use it for The user Uploaded

first image prediction uploads the successfully.
Autism image

2 User uploads Use it for The user Uploaded

first image prediction uploads the successfully.
non-Autism
image

CMRTC 40
 IDENTIFICATION OF AUTISM IN CHILDREN USING STATIC
FACIAL FEATURES AND DEEP NEURAL NETWORKS
6.3.2 CLASSIFICATION

Test case ID Test case name Purpose Input Output

Classification To check if the Autism image Predicted as

1 test 1 classifier is given. Autism.
performs its
task.

Classification To check if the Non-Autism Predicted as

2 test 2 classifier image is given. Non-Autism.
performs its
task.

CMRTC 41
 IDENTIFICATION OF AUTISM IN CHILDREN USING STATIC
FACIAL FEATURES AND DEEP NEURAL NETWORKS

7.CONCLUSION

CMRTC
 IDENTIFICATION OF AUTISM IN CHILDREN USING STATIC
FACIAL FEATURES AND DEEP NEURAL NETWORKS
7.CONCLUSION & FUTURE SCOPE

7.1 PROJECT CONCLUSION

In summary, our proposed system for detecting autism utilizes the strengths of
ResNet50 and Xception convolutional neural networks (CNNs) to analyze static facial
features, achieving impressive performance metrics. By prioritizing high accuracy
alongside well-balanced precision, recall, and F1-score, our method significantly
outperforms existing systems that rely on models such as MobileNet, VGG16, and YOLO.
The enhanced accuracy of our system is attributed to its advanced feature extraction
capabilities from 2D images, which are essential for recognizing subtle signs of autism
spectrum disorder (ASD).The combination of CNNs and deep neural networks (DNNs)
establishes our system as a dependable and effective solution for early autism detection.
This non-invasive, automated approach shows great potential for practical applications,
especially in the vital early screening processes for autism in children. By using high-
resolution facial images, our system effectively captures and analyzes distinct facial
features and expressions commonly linked to ASD.

In practical settings, this system can be easily implemented in hospitals, autism

centers, and pediatric clinics, providing a valuable resource for healthcare professionals. Its
user-friendly and non-invasive design makes it accessible for families seeking early
diagnoses for their children. The ability to swiftly and accurately screen for autism enables
timely intervention and support, which is crucial for enhancing developmental outcomes in
affected children.Moreover, the system's adaptability to various lighting conditions and
facial orientations improves its practicality, ensuring reliable performance in different
environments. By providing a sophisticated yet easy-to-use diagnostic tool, we aim to
connect clinical assessment with everyday practice, fostering early identification and
intervention for autism spectrum disorder.

CMRTC 42
 IDENTIFICATION OF AUTISM IN CHILDREN USING STATIC
FACIAL FEATURES AND DEEP NEURAL NETWORKS
7.2 FUTURE SCOPE

The future of our autism detection system, which utilizes ResNet50 and Xception
models, looks bright, with many opportunities for improvement. A key area to focus on is
the incorporation of diverse datasets that reflect different ethnicities, age groups, and
developmental stages. By training the model on a wider variety of data, we can improve its
ability to recognize a broader spectrum of autism-related facial features, resulting in more
inclusive and accurate diagnostics. Optimizing hyperparameters is crucial for enhancing
our model's performance. Making careful adjustments to factors like learning rate, batch
size, and architectural configurations can greatly boost the effectiveness of our CNN
models. Furthermore, using transfer learning techniques with pre-trained weights from
larger datasets can improve performance while reducing the need for extensive training.
An exciting possibility is the creation of real-time analysis capabilities. Moving from static
images to dynamic video input would enable continuous monitoring of facial expressions
and behaviors. This development could offer immediate feedback during assessments,
deepening our understanding of autism characteristics and allowing for ongoing patient
monitoring to facilitate timely interventions.

CMRTC 43
 IDENTIFICATION OF AUTISM IN CHILDREN USING STATIC
FACIAL FEATURES AND DEEP NEURAL NETWORKS

8.BIBILOGRAPHY

CMRTC
 IDENTIFICATION OF AUTISM IN CHILDREN USING STATIC
FACIAL FEATURES AND DEEP NEURAL NETWORKS
8. BIBILOGRAPHY
8.1 REFERENCES

[1] K.P.Lakshmi Priya, I.Jyothirmai, G.Akshaya,Ch. Smitha Chowdary. (2023).

IDENTIFICATION OF AUTISM IN CHILDREN USING STATIC FACIAL
FEATURES AND DEEP NEURAL NETWORKS.
https ://turcomat.org/index.php/turkbilmat/article/view/13702

[2] Reddy, P.; J, A. Diagnosis of Autism in Children Using Deep Learning

Techniques by Analyzing Facial Features. Eng. Proc. 2023, 59, 198.
https ://doi.org/10.3390/engproc2023059198

[3] Khan, B., Bhatti, S. M., & Akram, A. (2024). Autism Spectrum Disorder
Detection in Children Via Deep Learning Models Based on Facial Images.
Bulletin of Business and Economics (BBE), 13(1).
https ://doi.org/10.61506/01.00241

[4] Anjum, Jaasia and Hia, Naosin Akhter and Waziha, Anika and Kalpoma, Kazi
A.(2024). Deep Learning Based Feature Extraction from Children's Facial
Images for Autism Spectrum Disorder Detection.
https://doi.org/10.1145/3660853.3660888

[5] Akter, T.; Ali, M.H.; Khan, M.I.; Satu, M.S.; Uddin, M.J.; Alyami, S.A.; Ali,
S.; Azad, A.; Moni, M.A. Improved Transfer-Learning-Based Facial
Recognition Framework to Detect Autistic Children at an Early Stage. Brain
Sci. 2021, 11, 734.
https://doi.org/10.3390/brainsci11060734

[6] Ahsan, M.M.; Mahmud, M.A.P.; Saha, P.K.; Gupta, K.D.; Siddique, Z. Effect
of Data Scaling Methods on Machine Learning Algorithms and Model
Performance. Technologies 2021, 9,52.
https://doi.org/10.3390/technologies9030052

CMRTC 44
 IDENTIFICATION OF AUTISM IN CHILDREN USING STATIC
FACIAL FEATURES AND DEEP NEURAL NETWORKS

8.2 WEBSITES
[1] https://web.stanford.edu/class/cs231a/prev_projects_2016/output%20(1).pdf

CMRTC 45