Overview / Usage

The project has the ability to be Used on the edge across most of the devices and requires less time to get readily started and deployed because of its parallelized architechture, that helps it act like a real brain.

It can be Used across Websites ,IOT devices and other security solutions for easy and Fast Verification and prevent spoofing, A Further Usecase can be integreated upon this technology for being an accessible technology for the Blinds where they can have Glasses fitted with an camera and an earphone attached on their ears These types of glasses can help them Understand the surroundings, read books ,Feel emotions etc,whatever the Glasses see is reported to the blind person via a voice signal through the earphone attached.

Methodology / Approach

The first technique which comes to everyone's mind is Using the opencv Libraries and Using the LBPH and Voilla Jones Algorithm But even these algorithms fail under simplest Light conditions and fail to hold the bounding area within which the face resides.

Then Using next technology comes the Deep Learning which focuses on Convolutional Neural Networks for recognizing images, The first part comes into mind is using A Neural Net to Iterate through IMages and then use the CV2 Library and threshold the images for prediction on Images.But One eventually realizes that this cannot be holded true if we use a threshold every time i.e if the data set gets larger applying threshold for each and every person can become a tedious job.

​

SO we needed a New Solution.

It is a Indeed amazing to see how our brain can recognize Millions of Faces with their names and yet we didnt paid any attention to this fact.Now this is somewhat interesting isn't it, I will try explaing it somewhat simpler way how it actually does the job. All the faces we see are stored in some part of the brain that acts as a database. Whenever we see a face in front of eyes our eyes send signal to the brain and our brain tries to match the appearance of person from the list of appearances we have stored in the database of brain. If the brain is successful in seeing some sort of co-relation we Finally can tell if we have seen the person before or Not.

This Solution is Known as Siamese Neural Network

​

Which Uses two Neural Networks Parallely. One Neural Network Iterates through Images and the other Neural Network makes prediction on the images it sees. The second Neural Network Tries to Find Corelation from the Weights of The First Neural Network If the weights match to some extent we finally get the label

Technologies Used

List of Technologies Used:

• Keras(tensorflow-backend)
• Dell Precision Workstation
• Tensorflow
• Python
• Google Cloud
• OPencv Library
• Dlib 68-feature face recognizer

Repository

https://github.com/hr21/fast-facenet-at-oneshot