Trigeminal neuralgia is the most common cranial nerve disease. The domestic incidence rate is increasing these years. In the neuronavigation system, the registration and fusion of CT and MRA images can enrich the image content and guide doctors to better perform puncture operations. However, the registration accuracy is an important factor for the fused image. At present, many registration methods based on pixel similarity perform well in multi-modal brain images registration, but they generate local minima and have high computational complexity. Therefore, this paper proposes a multi-scale image registration method based on image pyramid with bilateral filter. Firstly, the rough registration of the affine transformation is performed on the images. Secondly, the fine registration of improved pyramid is carried out. The kernel function using the bilateral filter based on pyramid can better prevent the loss of edges and other detailed structures of the image. Thus, the image with low resolution in a pyramid can provide an effective initial value for high resolution images, and further improve the performance of the registration. The improved pyramid algorithm can effectively avoid local minima during the optimization of registration parameters, and alleviate the problem of uniformly smoothing. To evaluate the performance of the proposed method, this paper conducts experiment based on CT and MRA datasets. The experimental results show that the method in this paper achieves a more precise registration effect than image pyramid.