We are trying to build a deep model to recognize faces.
The dataset is stricted to 5 images for each face.
Five images for each (Trump, Putin, and Obama) are added as original images (our entire dataset).
Interesting approaches are utilized to make the approach functional.
Data augmentation increases the images in our dataset.
Image augmentation techniques used are scaling, cropping, flipping, padding, rotation, and translation.
Transfer learning provided us with a trained model to recognize faces.
The trained model used is VGG16.
We fine tuned the trained model to meet the specifications of our system.
Then we modified the models parameters by training it on our dataset.
finally, the model was tested on the test set and achieved 92.5% accuracy.
Other performance metrics (precision and recall) aren't taken into consideration becuase the dataset is balanced.
Moreover, the system was tested on new data too and it did well.
Codes are stored in notebooks file.
Original images folder contains the images taken from the internet without any processing.
Extracted Faces folder contains the faces extracted from the original images using Haar Cascade.
Augmented folder contains the augmented images.
Processed Folder contains splitted folders. Training, validation, and testing sets.