ASSIGNMENT ON

Research Paper Summerization

DIGITAL IMAGE PROCESSING

CSE 427

SUBMITTED TO
Naznin Sultana
Associate Professor
Department of CSE
Daffodil International University

SUBMITTED BY
SAKHWAT HOSAN
201-15-13731, Sec - A
Daffodil International University

Date of Submission
Fall 23
27 August 2023
1. Effective Computer Model For Recognizing Nationality From
Frontal Image

Bat-Erdene.B
Information and Communication Management School
The University of the Humanities Ulaanbaatar, Mongolia
e-mail: [email protected]

Ganbat.Ts
School of Information and Communication Technology The Mongolian
University of Science and Technology Ulaanbaatar, Mongolia
e-mail: [email protected]

The article introduces a computer model for identifying nationality from frontal
images using facial features.

Tools and Methods:

- Programming Languages: C++, C#, Matlab, Java.
- Libraries: OpenCV, EmguCV, Facelib, libsvm.
- Techniques: Haar-Cascade classifier for face detection, SVM for prediction,
AAM and ASM for face shape classification.

Dataset:
- Images from four countries: Mongolian, Japanese, Chinese, Korean.

Model Flow:
1. Face detection.
2. Image enhancement through edge detection.
3. Gender classification using PCA and Eigen face.
4. Face shape classification using AAM and ASM.
5. Feature extraction of facial attributes.
6. Anthropological measurements for distances and sizes.
7. SVM-based nationality prediction.

Accuracy:
- Experimentally achieved accuracy rates ranged from 78.3% to 86.4%.

Conclusion:
- The model effectively recognizes nationality from frontal images.
- Applications in military, police, defense, and immigration sectors suggested.
- Future work includes building a comprehensive database for improved
accuracy.
2. Nationality Detection using Deep Learning
Md. Abrar Hamim 1, Jeba Tahseen 1, Kazi Md Istiyak Hossain 1, and
Saurav Das 1
1 Department of Computer Science and Engineering Daffodil
International University,Dhaka, Bangladesh. 1*abrar15-
[email protected] [email protected] 1istiyakhossain15-
[email protected] [email protected]

The authors of the publication developed an intelligent model for gender and
nationality detection using deep learning techniques. They employed various
tools and methodologies to achieve their results.

Tools Used:
-Autocrop: A third-party library used for cropping frontal face images from the
dataset.
- CNN (Convolutional Neural Network): A type of deep learning architecture
specifically designed for image processing and recognition tasks.
- OpenCV and MS NET Framework: Utilized for facial recognition and feature
extraction in the project.
- Hybrid CNN and SVM Model: Used for abnormal incident detection in video
surveillance.

Dataset:
- The authors collected a dataset of frontal face images from individuals in
four countries: Bangladesh, Pakistan, India (Tamil), and Sri Lanka.
- This dataset served as the basis for training and testing the model's
accuracy in detecting gender and nationality.

Model Architecture and Complexity:

- The CNN model used convolutional, pooling, fully connected, and
normalisation layers to process the images.
- The model's architecture included a sequential structure with layers for face
detection, gender classification, and nationality recognition.
- The complexity of the model was reflected in the number of parameters, with
a total parameter count of 3,380,440. This includes both trainable and non-
trainable parameters.

Accuracy and Results:

- The model achieved an accuracy rate of up to 94.65% for nationality
detection and 55.97% for gender detection on the small dataset.
- Accuracy was measured based on the ratio of accurately predicted data
points to all data points.
3. Identifying Ethnics of People through Face Recognition: A
Deep CNN Approach
Ahmed Jawad A. AlBdairi ,1,2 Zhu Xiao ,1 and Mohammed Alghaili1
1College of Computer Science and Electronic Engineering, Hunan
University, Changsha 410082, China 2Computer Center, University
of Babylon, Hillah, Babil, Iraq

The research article titled "Identifying Ethnics of People through Face

Recognition: A Deep CNN Approach" presents a deep learning model for
recognizing the ethnicity of people based on their facial features.

Tools and Technologies Used:

Deep Learning: The research employs Convolutional Neural Networks
(CNNs), a popular deep learning architecture for image-related tasks.
Python: The programming language used for implementing the neural
network models and conducting the experiments.
TensorFlow or PyTorch: These are widely-used frameworks for building
and training deep learning models. The article doesn't explicitly mention
which framework was used, but these are commonly used choices.
GPU (Tesla K80): The training of deep neural networks often requires
significant computational resources. A GPU, such as Tesla K80, is
mentioned as the hardware used, likely to accelerate the training process.

Dataset:
Image Dataset: A new dataset containing facial images from three
different nationalities (Chinese, Pakistani, and Russian) was collected.
The dataset consists of 3141 images.
Data Collection: The images were collected from social media platforms
like Facebook and VK (a Russian social media website).

Model Architecture:
Convolutional Neural Network (CNN): The proposed model is based on
a CNN architecture with multiple convolutional and pooling layers.
Dropout Layers: Dropout layers are introduced within the network to
prevent overfitting. Two versions of the model are created: one with
dropout layers and one without.
Accuracy and Complexity:
Accuracy Comparison: The proposed model is compared with two state-of-
the-art models, VGG and Inception V3, which are well-known CNN
architectures.
Validation Accuracy: The proposed model achieved a validation accuracy
of 96.9% with regularization (dropout layers) and 96.6% without
regularization. This is the accuracy rate at which the model correctly
predicts the ethnicity of images it has never seen before.
Validation Loss: The validation loss is used as a metric to assess how well
the model generalizes. Lower validation loss indicates better
generalization.
Complexity: The article does not provide specific details about model
complexity. However, it's common for deep neural networks to have
millions of parameters, making them computationally intensive to train.

Overall, the study demonstrates the successful development of a deep

learning model for ethnicity recognition through facial features. The proposed
model outperforms other state-of-the-art architectures in terms of accuracy,
showcasing the potential of deep learning in solving complex image
recognition tasks.

4. A Classification of Arab Ethnicity Based on Face Image

Using Deep Learning Approach
NORAH A. AL-HUMAIDAN AND MASTER PRINCE Department of
Computer Science, Qassim University, Mulaydha 51452, Saudi
Arabia Corresponding author: Norah A. Al-Humaidan
([email protected]) This work was supported by the Qassim
University, Saudi Arabia, to complete Master Thesis under the
course M.S. in computer science

The research article titled "A Classification of Arab Ethnicity Based on Face
Image Using Deep Learning Approach" summary :

Tools and Techniques:

1.Convolutional Neural Networks (CNNs): CNNs are a type of deep learning
architecture particularly suited for image-related tasks due to their ability to
capture hierarchical features in images.
2. Deep Clustering:This is an unsupervised learning technique where the
model groups similar data points into clusters without labeled training data. It's
used here to explore ethnicity classification.
3. Pre-trained Models: The authors use a pre-trained CNN model called
ResNet-50. Pre-trained models are trained on large datasets and then fine-
tuned for specific tasks, allowing for better performance with limited labeled
data.
4. Data Pre-processing: The collected images are pre-processed to detect
and crop faces using the Dlib's pre-trained face detector. This ensures that
only relevant parts of the images (faces) are used for analysis.

Dataset:
The research involves the creation of an Arab image dataset. The dataset
includes images of individuals from different Arab sub-ethnic groups: Gulf
Cooperation Council (GCC) countries, the Levant, and Egypt. The images are
collected from publicly available sources, such as the internet, and are related
to specific professions (e.g., actors, singers, businesspeople). This dataset is
used to train and evaluate the classification models.

Models Used:
1. Supervised Learning (CNNs):The authors use a pre-trained ResNet-50
model for supervised learning. They fine-tune this model using various
hyperparameters (learning rate, optimizer, data augmentation, etc.) and
evaluate its performance on the Arab dataset. The goal is to classify the sub-
ethnic groups.

2. Unsupervised Learning (Deep Clustering):For unsupervised learning, the

authors apply deep clustering methods on the Arab dataset. They use three
different deep clustering methods: DEC (Deep Embedded Clustering), IDEC
(Improved Deep Embedded Clustering), and DynAE (Dynamic AutoEncoder).
These methods aim to group similar data points without predefined labels.

Accuracy:
1. Supervised Learning (CNNs):The accuracy achieved using the pre-trained
ResNet-50 model on the Arab dataset ranges from around 52.12% to 56.97%.
This means the model correctly classified the sub-ethnic groups in the images
with this level of accuracy.
2. Unsupervised Learning (Deep Clustering):The accuracy achieved using
deep clustering methods varies across experiments. It ranges from 32% to
59%. These methods achieve lower accuracy compared to the supervised
learning approach. The Normalized Mutual Information (NMI) and Adjusted
Rand Index (ARI) are also used to evaluate the quality of the clusters. NMI
and ARI scores indicate how well the clusters match the true labels; higher
scores are better.
5. A real time face emotion classification and recognition using
deep learning model
Dr. Shaik Asif Hussain, Ahlam Salim Abdallah Al Balushi
Department of Electronics and Communication Engineering, Middle
East College, Muscat.
Email: [email protected]

The research article titled "A real time face emotion classification and
recognition using deep learning model" summary :

Tools Used:

OpenCV: Used for facial detection using Haar cascade detection and
Viola-Jones algorithm.
Dlib: Utilized for face recognition, supporting face features using linear
SVM and Histogram of Oriented Gradients (HoG).
VGG Caffe Model: Employed for training and testing Convolutional Neural
Networks (CNNs) for face recognition and emotion classification.
Python Libraries: Included numpy, pandas, scikit-learn, and more for data
manipulation, computation, and machine learning tasks.
Anaconda: Used as the software platform for running the code and
implementing the system.
Graphical Processing Unit (GPU): Utilized for the intensive calculations
required by deep learning models.

Dataset:
KDEF Dataset: A dataset containing 4900 pictures, from which 1999 side-
profile pictures were excluded. The remaining 2901 images were split into
training and testing sets (70% and 30%, respectively).
Additional Images: The researchers also created their own dataset
containing 100 images.

Model Used:
Convolutional Neural Networks (CNNs): Specifically, the VGG16
architecture was employed for face recognition and emotion
classification. VGG16 is a deep CNN model known for its ability to
classify images with high accuracy.
Model Accuracy:
The paper reports that the designed system achieved an accuracy of
88% for face recognition and emotion classification using the CNN
model. This indicates a strong performance in accurately identifying
and classifying emotions based on facial expressions.
Conclusion:

The research paper presents a comprehensive system for real-time facial

detection, recognition, and emotion classification. By utilizing deep
learning techniques and tools such as OpenCV, Dlib, and VGG16, the
researchers developed a system that successfully detects human faces,
matches them with a database, recognizes emotions, and classifies them
accordingly.

The accuracy achieved by the system, at 88%, showcases its effectiveness

in identifying and classifying emotions from facial expressions. The paper
highlights the significance of the proposed system in various domains,
including education, industry, and medicine. It emphasizes the potential of
the system to aid in autism diagnosis, student emotion analysis in e-
learning, and more.