This paper proposes a novel image registration method based on the projection theorem of energy conservation, which can improve the performance of image registration. First, we build an inter-graph proximity matrix between nodes of a graph, and then construct intra-graph proximity measures for the individual node sets. The second stage involves the novel use of the projection theorem to project both the reference graph and the sensed graph into a lower dimensional feature space to reduce the dimensionality without losing any information of original data. Finally, we employ an analytical method rather than iterative approach in order to find the correct feature correspondences in the lower dimensional feature space between the graphs. Experiments on synthetic images and real-world images show the proposed method is effective and achieves high accuracy.