Recognition of Surgically Altered Face Image

Abstract

The use of plastic surgery for improving facial appearance is increasing rapidly, it involves altering a person's appearance through various techniques. It can also help those who have suffered physical injuries or have congenital deformities. Plastic surgery procedures can generally be categorized into two groups. Local plastic surgery aims to enhance specific features and improve skin texture, while global plastic surgery involves altering the overall structure and appearance of the face. This makes permanent changes in the structure of face which results in difficulties for recognition systems. While some individuals use these procedures to enhance their appearance, others may use them to conceal their identity, including those who have committed crimes, and due to this it becomes crucial to identify individuals accurately despite their altered appearance. The challenges posed by plastic surgery must be addressed to ensure accurate identification of individuals even after undergoing such procedures. Several studies have been conducted to tackle this problem, but it could be further improved by leveraging advanced deep convolutional neural networks. These networks can extract facial features that remain consistent before and after the procedure, which can be used to identify a person accurately. This research aims to overcome the challenges associated with recognizing individuals before and after plastic surgery. A convolutional neural network will be developed to extract facial features from pre- and post-surgery images, which will be used to train a model. Siamese network is developed to extract features from images. Siamese algorithm involves two identical neural networks that are trained on a set of images. After extracting features, the Siamese network compute distance between them. We will analyze the results obtained by Siamese network and custom CNN, then compare both of them.