An efficient approach is proposed for moving object segmentation in the H.264 compressed domain. The normalized motion vector field is derived from the H.264 compressed video and is exploited to segment the frame into motion-homogenous regions. The following region merging process is efficiently represented using a binary partition tree (BPT). Selective nodes are extracted from the traversal process of BPT to represent different moving objects. Experimental results demonstrate the good segmentation performance of our approach. In fact, the work presents a general BPT-based compressed-domain moving object segmentation approach, which can be easily adapted to other compressed domains including the Moving Picture Expert Group (MPEG) domain.