Face Detection Using Deep–Transfer Learning

Abstract

Face detection is a very important task to be performed in the field of computer vision and it has found its use in many applications like Biometrics, surveillance and in computer interaction with humans. Based on the literature it was found that Deep learning approaches for face detection are much more effective than other machine learning approaches [1-2]. Out of numerous models researchers have used VGG16[3] and AlexNet[4], which have been used very widely for face detection tasks using transfer learning.

The project starts by collecting our labelled images with and without face and some images where we are covering our face. We will augment these images to get more and variety images for our training. While augmenting we get flipped images, images with reduced and increased intensity of color providing us many different kind of cases to deal with while training the model.

We will compare the performance of VGG16 and AlexNet in our project for the task of face detection. VGG16 is a pretrained model which has been trained on a large dataset called ImageNet. VGG16 is primarily used to perform classification but here in our project we will fine tune it and we will perform regression to draw a bounding box around our face along with the classification task. Similarly, we will build an AlexNet model and perform regression and classification to detect and draw bounding box around our face.