Automatic Extraction of Lip Shape Using Feedforward Neural Network
Etsuro Nakamura, Tsuyoshi Takahashi, Yoichi Kageyama, Masaki Ishii, Makoto Nishida · IEEJ Transactions on Electronics Information and Systems · 2018
Lip motion features such as changes in the lip width and length provide important information for analyzing psychology and physical conditions. Furthermore, extracting lip motions from facial images of users has some advantages that are "needless of special equipment" and "possible of non-contact measuring". Our previous studies revealed that lip motions relate to mental state. Therefore, it is important to develop a method for extracting the lip shape from facial image data. However, the conventional method failed to extract lips due to individual differences of each subject and influences of the shadows on the periphery of lips. Lip extraction methods based on feature points need to use a lot of learning data. In this paper, we propose a novel lip extraction method using feedforward neural network (FFNN), which is one of machine learning methods. The proposed method is able to learn the features of lips using only one image, and its accuracy is same or higher than that of the conventional method and feature-point-based methods.