To address the problems that the infrared and visible images collected by distributed electro opticalimage system are susceptible to noise and poor alignment accuracy,an image registration algorithm based on convolutional neural network features and RIFT local features is proposed in this paper.Firstly,the convolution deep features of infrared and visible images to be registered are extracted based on improved AVIRnet,and the initial feature point pairs are obtained by deep features rough match.Secondly,the RIFT feature points are abstracted from overlap image area.Then,the local feature points are revised to obtain the final matching points pairs,and the accurate transformation matrix is estimated.The experimentresult indicated that by the two step registrations on deep and local features,the proposed method is invariant to nonlinear radiation difference and essentially satisfied the precise requirement of distributed electro optical infrared and visible image registration.