One-Shot Facial Image Recognition
- 0 Collaborators
The goal of this project is to build a One-Shot Face verification by collecting several images and saving them in the database as anchor images and then collect real-time images, both images are compared to give an output of either a True or False. ...learn more
Project status: Under Development
Groups
Student Developers for oneAPI
Intel Technologies
Intel Python
Overview / Usage
For a One-Shot Facial recognition, there is a database of K persons and then input images. The algorithm would check if the input image matched any of the K persons.
This project collects input images using the open-Source image processing framework – OpenCV. The images were detected and classified through the TensorFlow embedded layer of the Convolutional Neural Network, MaxPooling, and a Fully Connected Layer in an embedding Layer.
TensorFlow Binary Cross Loss was used to predict probabilities of the output to be either True or False (1 or 0). Then an Optimizer- ADAM was used to determine how my network would be updated based on the loss function
Methodology / Approach
Three data types were collected; wild-labeled Images(from different Individuals), An Anchor image which is my validation image, and a positive image(my input image to run the model).
The Anchor image and Positive image were collected using OpenCV, followed by an image augmentation of my anchor image and positive image, which were been processed through an embedding layer for detection and classification. L1 Distance Keras layer was used to calculate the similarities between the anchor image and the positive image followed.
The positive image was compared to the anchor image which gave a True value, while, then the anchor image was compared to the wild labeled faces (Negative Image) which gave a False value.
Finally, a Real-time verification was performed to know if my model was working by using an OpenCV to get a real-time image and then compare it to the anchor image, which came out True or False depending on the images that were collected and saved into the anchor images.
Technologies Used
OpenCV
Operating System Interface
NumPy
Matplotlib
TensorFlow
Python
Keras
Universal Unique Identifier (UUID)
My approach involves utilizing Intel's CPU's and GPU's via the DevCloud and ambassador program. Moreover, I aim to gain knowledge about Intel's oneAPI toolkits and libraries to enhance the efficiency of this project.
Repository
https://github.com/chelnnexy/One-Shot-Facial-Image-Recognition