We present an algorithm for fast global registration on colored dense point clouds. We combine CIELAB color information with fast point feature histograms for better correspondence and use color difference score to delete wrong pairs. The optimization is based on a method from Intel Labs that does not update correspondence or query closest-point during optimization. Experiments are taken on real partial overlapped data obtained from Kinect V2. The algorithm achieves a precise alignment with no initialization and cost little time. It can be extended to real-time robotics application for further study.