Automatic semantic propagation is an important technique to complete the image retrieval process which is provided to increase accuracy, speed and performance of image retrieval systems and also to decrease human expenses. In this paper, an automatic approach to propagate the semantics in the image data base is presented. The proposed approach provides a suitable framework for semantic propagation without user's interference at two different levels including local level and global level at the offline mode. This method propagates the related semantics of the image to its similar images based on visual similarity and semantic relations and also, updates the semantic relations between images and the semantic network. Evaluating results of our system provide the acceptable accuracy for semantic image retrieval after doing the automatic semantic propagation process.