Low-density Parity-Check (LDPC) code is widely used in distributed video coding (DVC) to realizing Wyner-Ziv (WZ) coding for its better performance than other channel codes. However, during the LDPC decoding process, most of DVC solutions do not make use of the video character to help decoding. In this paper, we propose an improved LDPC decoding scheme for DVC by utilizing the movement of the objects in the video sequence to partition different confidence areas for the side information (SI). The pixels in the SI correspond to different confidence areas, and the corresponding binary representations of the pixels will have different likelihood-ratio (LLR) initialization methods which are used to generate the LLR values for LDPC iterative decoding. The experimental results show that with the proposed decoding scheme the PSNR gain can reach up to 0.8 db without increasing the decoding complexity.