Facial Landmark Extraction Scheme Based on Semantic Segmentation

Abstract

Facial landmark is a set of features that can be distinguished in the human face with the naked eye. Typical facial landmark includes eyes, eyebrows, nose and mouth. It plays an important role in the human-related image analysis. For example, it can be used to determine whether human beings exist in the image, identify who the person is or recognize the orientation of a face when photographing. Methods for detecting facial landmark can be classified into two groups: One group is based on traditional image processing techniques such as Haar-cascade and edge detection. The other group is based on machine learning technique where landmark is detected through training facial features. However, such techniques have shown low accuracy, especially in the exceptional conditions such as low luminance or overlapped face. To overcome this problem, we propose a new facial landmark extraction scheme using deep learning and semantic segmentation and demonstrate that with even small dataset, our scheme can achieve excellent facial landmark extraction performance.