Distributed video coding (DVC) is a new kind of video codec scheme that exploits the statistical characteristics of source at the decoder part. The estimated side information is modeled with the original Wyner-Ziv frame and an additive noise. Side information plays an important role in determining the performance of the codec. In Turbo-or-LDPC-based DVC decoding, the log-likelihood ratio (LLR) for every source bit is calculated using side information and dependent channel model. In this paper, we propose a side information refinement method, which uses previously decoded bit-planes to update the side information for current bit-plane decoding. With the refined side information, more accurate channel model and bit LLRs are achieved, which results in a lower bit rate in video coding Rate-Distortion (RD) performance.