Multi-Channel Vibrotactile Signal Compression Based on Implicit Neural Representation
Yuto Ozawa, Akihiro Kuwabara, Tom Ogura, Takuya Fujihashi, Shunsuke Saruwatari, Takashi Watanabe · 2025
Haptic feedback is critical for enhancing immersion in extended reality applications. To enable realistic and immersive experiences, a compact representation of multi-channel vibrotactile signals is essential for storage and streaming. To obtain the compact representation, existing vibrotactile coding schemes leverage psychohaptic models to reduce perceptual redundancy in the haptic domain. However, the coding efficiency is still low. In this paper, we integrate the psychohaptic model and the implicit neural representation (INR), i.e., the power of neural networks (NNs), to obtain a further efficient representation for multi-channel vibrotactile signals. For this purpose, the proposed scheme overfits the multi-channel vibrotactile signals to a small NN and regards the overfitted weights of the NN as the representation of the vibrotactile signals. In addition, we propose a psychohaptic-inspired loss function for training the proposed NN architecture to obtain the compact representation with a slight degradation of the user's perceptual quality. Experiments using an open multi-channel vibrotactile dataset and the user perception-aware metric demonstrate that the proposed scheme simultaneously achieves compact and less quality-distorted vibrotactile representations compared with the state-of-the-art vibrotactile coding schemes.