Infrared-visual image registration is a common multi-sensor image processing task，which played an important role in the fields of remote sensing and military.In this paper,a novel and automatic method based on mutual information and gradient information is proposed for the infrared-visual image registration：Firstly,get the gradient information of the image，then the image edge can be obtained by using a structure tensor；the maximum normalized mutual information is used as similarity measure and also the Powell algorithm is used to search for the optimal registration parameters.Experimental results show that compared to the traditional method based on gradient and mutual information，the method proposed in this paper has higher accuracy and needs less time.It can be used as an effective method of coarse registration.