A self-adaptable image spam filtering system

Image spam embeds information in images to circumvent text-based spam-mail-filtering systems. Previous research does not consider cases in which the behavior of spammers changes over time. This study proposes a framework that can dynamically adapt to new types of image spam. The proposed framework is a two-layer imaging spam filtering system with a self-adaptable mechanism. The first layer is a fast classification module, which can filter many similar spam images very quickly. The second layer is a precise classification module, which classifies input images that are not readily classified by the first layer. Based on the proposed self-adaptable mechanism, the second layer immediately feeds spam image information back to the first layer. This allows the first layer to process new images using the updated information. Because the first layer quickly filters most spam images, this feedback approach improves system performance. This study reports the implementation of an example system based on the proposed framework. Experimental results show that the proposed system improves both accuracy and overall performance. Using limited training data, the proposed system achieved an accuracy of approximately 93.4%.