For channel estimation in the relay-assisted multiple-input multiple-output and orthogonal frequency division multiplexing indoor visible light communication (VLC) system affected by light source, there is a problem that the received signal cannot be separated by simple filtering. To solve this problem, a channel estimation by PARATUCK2-PARAFAC-based noise compensation which introduces tensors to construct the suitable model of VLC is proposed. First, the transmitting pilots for channel estimation and noise estimation are constructed separately with the Hermitian symmetry structure. Then the tensor-based transmission model of the indoor VLC system is established, and the signals for channel and noise estimations are received. Finally, an asymptotic estimation for separating signals from different sources and noise is proposed by PARATUCK2-PARAFAC. Simulation results show that the proposed algorithm provides a more effective scheme for channel estimation comparing with the state-of-the-art algorithms.