Abstract:
The purpose of this project is to develop a facial recognition algorithm having high
accuracy under rainy conditions. This report explores several techniques to solve the
challenges of rain artefacts, which are noise, blur and distortion of facial features in
images. A number of different stages of image processing are studied and discussed
including rain augmentation, rain removal, and model training. In this project, we
evaluated several methods in order to select the ID-CGAN for deraining. The rain
vector and blend generates rain streaks to augment images as noise to simulate rain.
Next, the ID-CGAN model is used for deraining, yet preserving the facial features,
while being aware that it should remove rain artifacts while not affecting facial
features. Then, two facial recognition models are trained using FaceNet technique
based on the augmented rainy images and the augmented images post deraining.
First, the rain artifacts are augmented on the dataset and then the ID-CGAN
deraining is applied on the image to enhance the image quality. It compares the
accuracy of the trained facial recognition models in recognising faces under rainy
conditions vs post deraining. This system is designed to be robust and reliable in the
areas of use such as security systems, surveillance and identity verification
