A Novel Method for Image Spam Filtering

Abstract

As spamming morphs from text-only to multi-media in content, the existing text-based spam filters are no longer effective for the task. Image analysis technology will provide an effective tool to solve the above problem. This paper proposes a novel filter approach that focuses on the spam which change the content text to image files, or embed the spam message into attached images. Spam images are analyzed and specific features are selected for inspection based on components-based method. Then the network junk-mail-filter system can use these features to identify whether to filter the mail or not by feature matching. Experiments show that our method is efficient in inspecting document images with different geometric structure. The proposed method has potential applications in document image information extraction and filtering.