In order to solve the problems like false detection and missing detection in the recognition results due to ignoring the relationship between different color components when using the image forgery detection algorithm to locate tampered content, the image forgery detection algorithm based on multivariate exponential moments and Euclidean locally sensitive Hash is proposed. The Gaussian low-pass filtering is introduced to eliminate the noise in suspicious images. Subsequently, the filtered image is segmented into a series of overlapping circular sub-blocks for improving its robustness to rotation and other content operations. Based on the Quaternion Exponential Moment(QEM), each circular sub-block is computed to extract the corresponding robust features and combine them into feature vectors. The corresponding Hash sequence of each sub-block is generated by the Euclidean locally sensitive hashing mechanism. The spatial distance between any two adjacent Hash elements is computed, and the matching of all sub-blocks is finished by comparing with the preset threshold. Finally, by means of the consistency method of random samples, the false matching is eliminated, and the tampered content is located by morphological processing. The experimental results show that compared with the existing forgery detection methods, the proposed algorithm has higher accuracy of forgery detection under various geometric modifications.