Shadow Removing Method for Lip Shape Extraction Using Feedforward Neural Network
Etsuro Nakamura, Tsuyoshi Takahashi, Yoichi Kageyama, Masaki Ishii, Makoto Nishida, Motonari Shirasu · 2019 IEEE 1st Global Conference on Life Sciences and Technologies (LifeTech) · 2019
It was clarified that there is a relationship between lip motion during utterance and amusement feelings. To use in daily life, it is important to extract the minute changes in lip motion during utterances. Therefore, the lip shape extraction method (previous method) using a feedforward neural network was proposed and its usefulness was indicated. However, the previous method was affected by the shadow of the lower part of the lip and thus excessively extracts the lip area. In this paper, a new method was proposed combining a shadow removing method and the previous method to correct the excessive extraction. Lips for utterance images of seven subjects were extracted using the proposed method and the previous method, and each Intersection over Union (IoU) was calculated using lip extraction results. Compared with IoU, the proposed method improves the IoU by 0.0342 (on average) and the average of IoU was 0.8812.