Gender Recognition Through Face Using Deep learning

Project Title

Gender Recognition Through Face Using Deep learning

Project Area of Specialization Computer ScienceProject Summary

Abstract

Humans are capable of determining an individual’s gender relatively easily using facial attributes. Although it is challenging for machines to perform the same task, in the past decade incredible strides have been made in automatically making prediction from face image. The project identifies or detects the gender from the given face images. The tools used involve Convolutional Neural Network along with programming language like Matlab. The project has been motivated by problems like lack of security, frauds, child molestation, robbery, criminal identification. Keywords: CNN, Gender, Machine Learning, Matlab, Deep Learning.

Clear Statement of the Problem

The conventional techniques in the compute vision domain requires an accurate information about facial landmarks but for complex images, particularly in the real-time captured videos, it is not possible to get these face landmarks information as per static images. Therefore, a few existing techniques performed some preprocessing techniques to improve the initial information (static image). After these operations, the researcher’s extracted shape and local features for better accuracy but it is only depending on the expertise of the researchers to get the most optimal solutions (features). In the real-time video sequences, it is not possible to combine these features because number of steps increase the system computational time.  

Project Objectives

Objectives

The main objective of this project is to create an efficient system for gender recognition using deep learning. Thus, the goals below can be derived from this:

· The system should be able to extract features from the entire video frame.

· The system should be able to classify an individual based on his/her gender.

· To be executed in less computational time.

· Compared with existing deep learning models for better evaluation.

Motivation

· The number of crimes has been increasing daily at a much faster rate. It has become a necessity to identify criminals as soon as possible. The traditional way of identification is a slow process while the proposed approach can be used to counter terrorism by identifying the features at a much faster rate.

· The project can also be used to overcome the frauds that can take place during voting i.e. can be used for voter identification.

Project Implementation Method

Introduction

Gender Face detection is an easy and simple task for humans, but not so for computers. It has been regarded as the most complex and challenging problem in the field of computer vision due to large intra-class variations caused by the changes in facial appearance, lighting and expression. Such variations result in the face distribution to be highly nonlinear and complex in any space that is linear to the original image space.

Face detection is the process of identifying one or more human faces in images or videos. It plays an important part in many biometric, security and surveillance systems, as well as image and video indexing systems.

Gender recognition of face images is an important task in computer vision as many applications depend on the correct gender assessment. Examples of these applications include visual surveillance, marketing, intelligent user interfaces, demo graphic studies, etc. The gender recognition problem is usually divided into several steps, similarly to other classification problems object detection, preprocessing, feature extraction and classification. In the detection phase, the face region is detected and cropped from the image. Then, a preprocessing technique is used to reduce variations in scale and illumination. After this normalization, the feature extraction step aims at obtaining representative and discriminative descriptors of the face region.

Finally, a binary classifier that learns the differences between male and female representations is trained.

Feature extraction is the most critical step in order to achieve good performance. Traditionally, features have come up as a result of the knowledge and expertise of many feature practitioners. However, instead of relying on this human-based process to define the best representation of the data in a specific problem, it would be much more interesting to let the algorithm to discover that representation automatically by itself.

Our model extracts several local features from the input images, and these features feed a discriminative deep neural network. The network learns to classify each local feature according to the label of the image to which it be longs. The final decision for the whole input image is taken based on a simple voting scheme that takes into account all the local contributions. We have found that for some specific applications, where some registration has been applied to the images, e.g. a face detector, the Local-DNN has demonstrated to be superior to other techniques due to a greater robustness to small translations, occlusions and local distortions.

Benefits of the Project

FUTURE WORKS:

Upon changing the dataset, the same model can be trained to predict emotion, age, ethnicity, etc. The gender classification can be used to predict gender in uncontrolled real time scenarios such as railway stations, banks, bus stops, airports, etc. For example, depending upon the number of male and female passengers on the railway station, restrooms can be constructed to ease the travelling.

Applications

• School surveillance cameras
• Banks and residential areas
• Voting
• Data flow graphs
• Suspects in crimes

Technical Details of Final Deliverable

Project Plan / Schedule

• To start with the project, the first step that needs to be done is data collection.

Datasets play an important in deep learning as it is used to train the system to get the required output. This dataset has images of people of different gender.

• Training Dataset into frames

• Normalization

• The model is trained using ConvNet (Convolutional Neural Network) consisting of 5 layers. The CNN consists of many hidden layers such as Convolutional layer, Relu layer, Max Pooling layer, Fully Connected layer.

•Transfer layers

• Test Dataset

• Model training

• Using these layers, the input face image is converted into weights and saved in ‘.h5’ format. These weights are then used to predict an unknown image. The average accuracy achieved in the project is 90%

Final Deliverable of the Project HW/SW integrated systemCore Industry SecurityOther IndustriesCore Technology Artificial Intelligence(AI)Other TechnologiesSustainable Development Goals Gender Equality, Sustainable Cities and CommunitiesRequired Resources

Item Name	Type	No. of Units	Per Unit Cost (in Rs)	Total (in Rs)
			Total in (Rs)	77000
Sony camera ?850,	Equipment	1	58000	58000
Gpu Asus Radeon RX 550 2G GDDR5 HDMI DVI AMD	Equipment	1	11000	11000
For connections	Miscellaneous	1	8000	8000