Abstract: In order to avoid the detection of the spam system based on text, more and more spammers have embedded text information into the image. An image spam identification method based on gray-gradient co-occurrence matrix (GGCM) was proposed to detect image spam effectively. The feature of image was extracted through GGCM firstly, and then LS-SVM was used to do classification. The test results show that this method has higher classification accuracy and better real-time performance．