Many infrastructure networks interact with and depend on each other to provide proper functionality. The interdependence between networks has catastrophic effects on their robustness. Events taking place in one system can propagate to any other coupled system. Recently, great efforts have been dedicated to the research on how the coupled pattern between two networks affects the robustness of interdependent networks. However, how to dynamically construct the links between two interdependent networks to obtain stronger robustness is rarely studied. To fill this gap, a global homogenizing coupled pattern between two scale-free networks is proposed in this paper. Making the final degrees of nodes distributed evenly is the principle for building the dependency links, which has the following two merits. First, the system robustness against random failure is enhanced by compressing the broadness of degree distribution. Second, the system invulnerability against targeted attack is improved by avoiding dependence on high-degree nodes. In order to better investigate its efficiency on improving the robustness of coupled networks against cascading failures, we adopt other four kinds of coupled patterns to make a comparative analysis, i.e., the assortative link (AL), the disassortative link (DL), the random link (RL) and global random link (GRL). We construct the BA-BA interdependent networks with the above 5 coupled patterns respectively. After applying targeted attacks and random failures to the networks, we use the ratio of giant component size after cascades to initial network size to measure the robustness of the coupled networks. It is numerically found that the interdependent network based on global homogenizing coupled pattern shows the strongest robustness under targeted attacks or random failures. The global homogenizing coupled pattern is more efficient to avoid the cascading propagation under targeted attack than random failure. Finally, the reasonable explanations for simulation results is given by a simply graph. This work is very helpful for designing the interdependent networks against cascading failures.