Multimodal template matching based on gradient and mutual information using scale-space