- As an individual project for my Summer 2018 Internship at the Port of Seattle as a Machine Learning & Visual Data Intern, I implemented a Facial Detection & Recognition system. In this project, I heavily utilized Python and Jupyter Notebook to run my Python code. In addition, I also used OpenCV to get a lot of the face detectors as well as a recognizer. A really neat and interesting feature to this implementation is that it is able to collect and train new data almost instaneously. In other words, it is able to collect new training data based on the specifications from a couple of very basic user inputs.
- Note: This is not the best implementation there is. Of course, there are deep learning models out there that will produce much more accurate results. However, what makes this specific implementation attractive is the time that it takes to compute and train the data; maximum a couple of minutes with ~300 images. On the other hand, other deep learning models that utilize neural networks will take much more time and are a lot more computationally expensive.
- My future goal moving on with Facial Detection & Recognition is to definitely experiment and train with deep learning models. Getting the best predictions is certainly something very viable to aim for.
- I hope you enjoy this implementation of a Facial Detection & Recognition system that I was able to come up with. Please contact me if you have any suggestions or comments!
- The "Barack Obama" folder is for the first training, as the system needs images in order to "train" on. So, the "Barack Obama" folder is just a starting set of images. If you wish, you can always delete the folder and add a folder of existing images of your own preference.
- https://docs.opencv.org/2.4/modules/contrib/doc/facerec/facerec_tutorial.html
- https://www.superdatascience.com/opencv-face-recognition/
- https://towardsdatascience.com/face-recognition-how-lbph-works-90ec258c3d6b
- http://www.scholarpedia.org/article/Local_Binary_Patterns
