FACE SHAPE CLASSIFIER USING DEEP LEARNING
0 likes68 views
This document presents a system that uses deep learning to classify human faces into five different shape categories (heart, oval, square, oblong, circle) with an accuracy of 82%. The system takes in facial images and uses a pre-trained convolutional neural network (CNN) model to predict the best matching face shape. It is designed to help with tasks like recommending hairstyles, frames, and clothing based on a person's face shape. The CNN model is trained on a dataset of faces, validated on a separate set, and tested for its performance in classifying new images into the five shape categories.
![International Research Journal of Engineering and Technology (IRJET) e-ISSN: 2395-0056
Volume: 09 Issue: 12 | Dec 2022 www.irjet.net p-ISSN: 2395-0072
© 2022, IRJET | Impact Factor value: 7.529 | ISO 9001:2008 Certified Journal | Page 361
FACE SHAPE CLASSIFIER USING DEEP LEARNING
Rohan S1, Suhas R Vittal2, Neeraj H Gowda3, Dr. Priya R Sankapal4
1,2,3 Student, Dept. of Electronics and Communication, BNMIT, Bangalore, Karnataka, India
4 Assistant Professor, Dept. of Electronics and Communication, BNMIT, Bangalore, Karnataka, India
---------------------------------------------------------------------***---------------------------------------------------------------------
Abstract - Beauty and cultural activities, such as
hairstyles, clothes and accessories, and other costumes,
demand an understanding of a person's face shape, which is
not always accurate or efficient and requires specialist
expertise. In this paper, we presenta computer-aidedsystem
based on image processing and deep learning that can
automatically recognize the facial form without the need for
an expert. As input, the system receives face photos and
delivers them to a pre-trained model. This algorithm
classifies and predicts outcomes based on face shape across
five shapes: heart, oval, square, oblong, and circle, and
provides results for the best match with an accuracyof82%.
Key Words: Deep Learning, K-Nearest Neighbors, Linear
Discriminant Analysis, Support Vector MachineswithLinear
Kernel, Convolutional Neural Network.
1. INTRODUCTION
Our faces are our first and last impressions. Thus, we must
learn to leverage this. Understandingourfaceshapehelpsus
enhance our potential. We can highlight or hide features
with a little expertise in this area. By determining our face
shape, we can learn how to style ourselves to look appealing
and natural rather than unpleasant and boring. Finding our
face form assists our style. Our outfit choices are benefited
from our face structure.
Re-trainable custom image classifiers make face shape
classifiers easier to design. The classifier should be capable
of recognizing whether a frontal view image of a human face
is heart, oblong, oval, round, or square. Fashion stylists
recommend frames and hairstyles basedonfaceform.OPSM
Opticians advocate bigger frames for oblong-shapedfaces to
balance long and wide features and avoid transparent
rimless frames that exaggerate length and width. Similar
apps recommend hats, makeup, jewelry, and other fashion
accoutrements.
These recommendation algorithms could be part of a bigger
personal digital assistant linked to social media and product
advertisements. Algorithms could potentially advise virtual
or cosmetic facial changes to improve one's appearance.
Face-shape categorization systems can speed up facial
recognition, but more abstracted profiling schemes
employing system-learned classes may be better. Online
guides, apps, and mobile apps are available face shape
classifiers in the literature. Two scientific articles are peer-
reviewed. Publishedfaceshapeclassificationmethodscollect
pre-defined features from images and train classifiers using
these methods: K-Nearest Neighbors (KNN), Linear
Discriminant Analysis (LDA), Support Vector Machineswith
Linear Kernel (SVMLIN), and SVM-RBF [2]. (MLP). While
published overall accuracies of 64.2% to 85.0% look good, it
would be interesting to compare these results with a
classifier employing convolutional neural networks (CNNs),
which are becoming moreprominentinpictureclassification
challenges.
1.1 MOTIVATION
Having prior knowledgeaboutourownfacialshapeshelpsus
make better choices about appareland fashion. Our abilityto
style ourselves to look our best without consulting an expert
on how our clothingchoices relateto ourfaceshapeishelped
by the coherence of our face shape and the fashion choices
that we choose.
1.2 OBJECTIVES
This project aims at developing a neural network to
recognize a person's face shape from facial photos and
present the results. This approach is best for face shape
identification at opticians, beauty salons, and at home. Our
goal is to help consumers find the perfect style for their face
shape. This allows women choose from a wide variety of
beautiful clothes, hairstyles, and makeup for different
occasions. This projectcanbeimplementedintoe-commerce
platforms to assist us choose clothes that suit our facial
shapes and eliminate items that don't. Face shape
classification helps us choose products in all areaswhere we
shop.
2. LITERTURE SURVEY
Face recognition, one of the most successful image analysis
and comprehending applications, has garnered attention in
recent years. After 30 years of research, feasible technology
and a wide range of commercial and law enforcement
applications explain this tendency. Machine recognition
systems have matured, yet many real applications limit their
success. Face identification in outdoor photographs with
changing illumination and position is still a major challenge.
Thus, current systems cannot match human vision. Face
shapedetection, another spinoff from facerecognition,offers
a wide range of applications in businesses that cater to
individuals, such as fashion and e-commerce. We referenced](https://image.slidesharecdn.com/irjet-v9i1262-230114100805-0669f3d9/85/FACE-SHAPE-CLASSIFIER-USING-DEEP-LEARNING-1-320.jpg)
![International Research Journal of Engineering and Technology (IRJET) e-ISSN: 2395-0056
Volume: 09 Issue: 12 | Dec 2022 www.irjet.net p-ISSN: 2395-0072
© 2022, IRJET | Impact Factor value: 7.529 | ISO 9001:2008 Certified Journal | Page 363
Fig -4: Accuracy of the model
After training for up to 10 epochs, the overall accuracyofthe
face shape classifier algorithm is around 82%. The model’s
accuracy can be further increased by performing further
epochs, resulting into better performance of the algorithm.
Fig -5: Confusion Matrix
Using a confusion matrix, we evaluate the performance of
our model. This is frequently used to describe how a
classification model, also known as a "classifier," performed
on a set of test data for which the true values were known.
5. CONCLUSIONS
This application reads a facial photograph as input and
provides the most accurate prediction of the face shape. The
outcomes of each image analysis show that the system is
trained, validated, and tested. The built system can
accurately determine facial form. There is, nevertheless,
room for some prospective improvements. This system will
be trained on 5000 faces of women ranginginagefrom20to
40 years to recognize shapes such as square, oval, oblong,
round, and heart. This technique can be used to develop
applications such as eyeglasses, haircuts, and apparel to
provide a nice feeling of fashion.
REFERENCES
[1] SV Rajapaksha and BTGS Kumara "Hairstyle
Recommendation Based on Face Shape Using Image
Processing”.
[2] H. N. Gunasinehe and T.G.I Fernanda "Identifying the
Face shape of a Person using Machine Learning
Techniques”.
[3] Pallab P. "Face It – The Artificially Intelligent
Hairstylist".
[4] Adonis Emmanuel DC. Tio “Face shape classification
using Inception v3”.
[5] Amir Zafar and Tiberiu Popa. 2016. Face and Frame
Classification using Geometric Features for a Data-
driven Frame Recommendation System. In Proceedings
of the 42nd Graphics Interface Conference (GI '16).
Canadian Human Computer Communications Society,
School of Computer Science, University of Waterloo,
Waterloo, Ontario, Canada, 183-188.
[6] W. Sunhem and K. Pasupa, "An approach to face shape
classification for hairstyle recommendation," Eighth
International Conference on Advanced Computational
Intelligence (ICACI), Chiang Mai.
[7] Eshwar S G, Gautham Ganesh Prabhu J, A. V. Rishikesh,
Charan N A and Umadevi V, "Apparel classificationusing
Convolutional Neural Networks," International
Conference on ICT in Business Industry & Government
(ICTBIG), Indore.](https://image.slidesharecdn.com/irjet-v9i1262-230114100805-0669f3d9/85/FACE-SHAPE-CLASSIFIER-USING-DEEP-LEARNING-3-320.jpg)
FACE SHAPE CLASSIFIER USING DEEP LEARNING
- 1. International Research Journal of Engineering and Technology (IRJET) e-ISSN: 2395-0056 Volume: 09 Issue: 12 | Dec 2022 www.irjet.net p-ISSN: 2395-0072 © 2022, IRJET | Impact Factor value: 7.529 | ISO 9001:2008 Certified Journal | Page 361 FACE SHAPE CLASSIFIER USING DEEP LEARNING Rohan S1, Suhas R Vittal2, Neeraj H Gowda3, Dr. Priya R Sankapal4 1,2,3 Student, Dept. of Electronics and Communication, BNMIT, Bangalore, Karnataka, India 4 Assistant Professor, Dept. of Electronics and Communication, BNMIT, Bangalore, Karnataka, India ---------------------------------------------------------------------***--------------------------------------------------------------------- Abstract - Beauty and cultural activities, such as hairstyles, clothes and accessories, and other costumes, demand an understanding of a person's face shape, which is not always accurate or efficient and requires specialist expertise. In this paper, we presenta computer-aidedsystem based on image processing and deep learning that can automatically recognize the facial form without the need for an expert. As input, the system receives face photos and delivers them to a pre-trained model. This algorithm classifies and predicts outcomes based on face shape across five shapes: heart, oval, square, oblong, and circle, and provides results for the best match with an accuracyof82%. Key Words: Deep Learning, K-Nearest Neighbors, Linear Discriminant Analysis, Support Vector MachineswithLinear Kernel, Convolutional Neural Network. 1. INTRODUCTION Our faces are our first and last impressions. Thus, we must learn to leverage this. Understandingourfaceshapehelpsus enhance our potential. We can highlight or hide features with a little expertise in this area. By determining our face shape, we can learn how to style ourselves to look appealing and natural rather than unpleasant and boring. Finding our face form assists our style. Our outfit choices are benefited from our face structure. Re-trainable custom image classifiers make face shape classifiers easier to design. The classifier should be capable of recognizing whether a frontal view image of a human face is heart, oblong, oval, round, or square. Fashion stylists recommend frames and hairstyles basedonfaceform.OPSM Opticians advocate bigger frames for oblong-shapedfaces to balance long and wide features and avoid transparent rimless frames that exaggerate length and width. Similar apps recommend hats, makeup, jewelry, and other fashion accoutrements. These recommendation algorithms could be part of a bigger personal digital assistant linked to social media and product advertisements. Algorithms could potentially advise virtual or cosmetic facial changes to improve one's appearance. Face-shape categorization systems can speed up facial recognition, but more abstracted profiling schemes employing system-learned classes may be better. Online guides, apps, and mobile apps are available face shape classifiers in the literature. Two scientific articles are peer- reviewed. Publishedfaceshapeclassificationmethodscollect pre-defined features from images and train classifiers using these methods: K-Nearest Neighbors (KNN), Linear Discriminant Analysis (LDA), Support Vector Machineswith Linear Kernel (SVMLIN), and SVM-RBF [2]. (MLP). While published overall accuracies of 64.2% to 85.0% look good, it would be interesting to compare these results with a classifier employing convolutional neural networks (CNNs), which are becoming moreprominentinpictureclassification challenges. 1.1 MOTIVATION Having prior knowledgeaboutourownfacialshapeshelpsus make better choices about appareland fashion. Our abilityto style ourselves to look our best without consulting an expert on how our clothingchoices relateto ourfaceshapeishelped by the coherence of our face shape and the fashion choices that we choose. 1.2 OBJECTIVES This project aims at developing a neural network to recognize a person's face shape from facial photos and present the results. This approach is best for face shape identification at opticians, beauty salons, and at home. Our goal is to help consumers find the perfect style for their face shape. This allows women choose from a wide variety of beautiful clothes, hairstyles, and makeup for different occasions. This projectcanbeimplementedintoe-commerce platforms to assist us choose clothes that suit our facial shapes and eliminate items that don't. Face shape classification helps us choose products in all areaswhere we shop. 2. LITERTURE SURVEY Face recognition, one of the most successful image analysis and comprehending applications, has garnered attention in recent years. After 30 years of research, feasible technology and a wide range of commercial and law enforcement applications explain this tendency. Machine recognition systems have matured, yet many real applications limit their success. Face identification in outdoor photographs with changing illumination and position is still a major challenge. Thus, current systems cannot match human vision. Face shapedetection, another spinoff from facerecognition,offers a wide range of applications in businesses that cater to individuals, such as fashion and e-commerce. We referenced
- 2. International Research Journal of Engineering and Technology (IRJET) e-ISSN: 2395-0056 Volume: 09 Issue: 12 | Dec 2022 www.irjet.net p-ISSN: 2395-0072 © 2022, IRJET | Impact Factor value: 7.529 | ISO 9001:2008 Certified Journal | Page 362 to papers and publications on face classifications, a mix of face recognition and image classification. Our survey showed that CNN was best for our strategy because it found and clustered features of each category without training. We conducted this study to give an up-to- date review of face classification literature and to shed light on machine face recognition research. 3. METHODOLOGY Input Image Fig -1: Block Diagram The data set is divided into training andtestingsets;theCNN model is trained using the training sets, and then tested on the testing data setstoanalyzeitsperformance.Performance metrics such as accuracy are assessed among the five categories specified above. Convolutional Neural Network (CNN) based machine learning algorithms have been created to automatically analyze face appearance-based difficulties. The issue of feature alignment can be solved using CNNs, and experimental results show that they perform well on challenging test images.Asystemthatemploysa hierarchical process in which it first approximatelyidentifiestheposition of the eyes, nose, and mouth before improving the outcome through the detection of various facial feature points. A proposed face recognition system uses a particular CNN architecture that enables nonlinear mapping of the image's area into a lower dimensional subspace for better classification. The technique was used on numerous public face databases and was able to outperform traditional face recognition logics in terms of recognition rate. Users can use this dataset to train and test the accuracy of their own convolutional neural networksbecauseitcontains thousands of images of handwritten numbers. Many people have produced and gathered a variety of well-known datasets to aid in the development and study of convolutional neural networks. 4. Results Fig -2: Output of the Algorithm Fig -2 demonstrates the square shape of the input image's outcome, as seen in the image. The CNN algorithm classifies the image into 5 possible shapes and then predicts which shape is the most appropriate. Fig -3: Output of the Algorithm Validation Set Training Set CNN Model Performance Analysis Square Circle Oblong Oval Heart
- 3. International Research Journal of Engineering and Technology (IRJET) e-ISSN: 2395-0056 Volume: 09 Issue: 12 | Dec 2022 www.irjet.net p-ISSN: 2395-0072 © 2022, IRJET | Impact Factor value: 7.529 | ISO 9001:2008 Certified Journal | Page 363 Fig -4: Accuracy of the model After training for up to 10 epochs, the overall accuracyofthe face shape classifier algorithm is around 82%. The model’s accuracy can be further increased by performing further epochs, resulting into better performance of the algorithm. Fig -5: Confusion Matrix Using a confusion matrix, we evaluate the performance of our model. This is frequently used to describe how a classification model, also known as a "classifier," performed on a set of test data for which the true values were known. 5. CONCLUSIONS This application reads a facial photograph as input and provides the most accurate prediction of the face shape. The outcomes of each image analysis show that the system is trained, validated, and tested. The built system can accurately determine facial form. There is, nevertheless, room for some prospective improvements. This system will be trained on 5000 faces of women ranginginagefrom20to 40 years to recognize shapes such as square, oval, oblong, round, and heart. This technique can be used to develop applications such as eyeglasses, haircuts, and apparel to provide a nice feeling of fashion. REFERENCES [1] SV Rajapaksha and BTGS Kumara "Hairstyle Recommendation Based on Face Shape Using Image Processing”. [2] H. N. Gunasinehe and T.G.I Fernanda "Identifying the Face shape of a Person using Machine Learning Techniques”. [3] Pallab P. "Face It – The Artificially Intelligent Hairstylist". [4] Adonis Emmanuel DC. Tio “Face shape classification using Inception v3”. [5] Amir Zafar and Tiberiu Popa. 2016. Face and Frame Classification using Geometric Features for a Data- driven Frame Recommendation System. In Proceedings of the 42nd Graphics Interface Conference (GI '16). Canadian Human Computer Communications Society, School of Computer Science, University of Waterloo, Waterloo, Ontario, Canada, 183-188. [6] W. Sunhem and K. Pasupa, "An approach to face shape classification for hairstyle recommendation," Eighth International Conference on Advanced Computational Intelligence (ICACI), Chiang Mai. [7] Eshwar S G, Gautham Ganesh Prabhu J, A. V. Rishikesh, Charan N A and Umadevi V, "Apparel classificationusing Convolutional Neural Networks," International Conference on ICT in Business Industry & Government (ICTBIG), Indore.