Disclosed herein are a feature point used to localize an image-based robot and build a map of the robot and a method of extracting and matching an image patch of a three-dimensional (3D) image, which is used as the feature point. It is possible to extract the image patch converted into the reference image using the position information of the robot and the 3D position information of the feature point. Also, it is possible to obtain the 3D surface information with the brightness values of the image patches to obtain the match value with the minimum error by a 3D surface matching method of matching the 3D surface information of the image patches converted into the reference image through the ICP algorithm.