Compressing code is known to be one of the most effective methods in reducing the energy consumed in the interface between memory and processor. In this paper, we addressed the problem, which has never been tackled in the previous code compression techniques, of determining binary code of the instructions to be compressed. Our observation is that a careful assignment of binary code to the instructions to be compressed can lead to a considerable amount of power saving since the switching activity in instruction accesses varies significantly depending on the ways of assigning binary code to the instructions. To achieve the power saving, we analyze the problem and transform it into a graph optimization problem and solve it efficiently, but effectively by inventing an incremental node covering technique. From experiments using a set of benchmark programs, it is shown that the approach is quite effective, producing code with 17%-42% less switching activity in instruction accesses over a greedy low-power binary code assignment.