A matching algorithm based on local feature descriptors is presented to solve the difficulties of IR/visual image matching.The matching process is accomplished by three steps:Firstly,the control points are extracted by the Difference-of-Gaussian corner detecting algorithm.Secondly,the statistical information of gradient direction which has been connected with gradient mirroring method is calculated,and the shape context descriptor is introduced.Then a joint descriptor combining with SIFT and shape context is applied to describe points.Finally,the matched distance of joint descriptors is obtained with the weighted average integration of Euclidean distance for SIFT and Chi square distance for Shape Context.For large number of infrared and visible images,the experimental results demonstrate that the method is more effective than the original SIFT with a higher average correct matching rate.