Template matching is crucial and difficult in object detection and image processing. Many current template matching algorithms often lack robustness when facing illumination changes, partial occlusions, and cluttered backgrounds, resulting in low accuracy of matching results. To enhance the precision of template localization, we propose an accuracy method based on a fully convolutional Siamese network for template matching, transforming the template matching task into a center point estimation task. We design a novel Siamese network backbone for effectively extracting feature information from input image pairs, improving the robustness to scale variations while reducing the computational complexity. We use the center point to represent the object corresponding to the template on the search image, completely discarding the estimation of other attributes of the object. Extensive experiments on standard template matching datasets demonstrate our method achieves state-of-the-art performance.