Image Spam Detection Mechanism And Algorithm

Image spam is a new type of spam. Spammers embed information into image sent by e-mail, in order to avoid text-based spam filtering system detected. The detection of image spam is one of the hot research area in internet spam detection. The goal of detect image spam is to solve the current spam filters can’t effectively filter image spam.The image which is embedded into image spam is the object of our study. Firstly, this paper analyzes the feature of image spam. Then, we summarize and analyze the currently used method of filtering spam image. In this paper, we improve the following four aspects of current method.1. In Near-duplicate detection, most researcher use SIFT algorithm and SURF algorithm. Although these algorithm achieve great result, but the efficiency of these two algorithms is relatively low. In this paper, BRISK algorithm is used in the area of image spam detection which can significantly improve the efficiency of detection.2. Artificial neural network (ANN) algorithm is use in image spam detection area in this paper. Since ANN algorithm for linearly inseparable problem has better classification results, and also has some self-learning, adaptive ability, so this algorithm can meet the ever-changing spam image detection problems.3. Text area features are the most significant features in detect image spam, so accurately extract text area from image is the key problem. Based on the noise features of image spam, we present a small noise removal algorithm, which can more accurately extract the text area from image spam.4. Since the classification algorithm and Near-duplicate detection algorithms have some shortcomings. This paper proposes a progressive image spam detection mechanism, which is a two-layer detection mechanism. The first layer of this mechanism filter image spam by using near-duplicate detection, the second layer is formed using classification algorithm filtering image spam. And the detection result of the second layer can feed back to the first layer, to improve the detection capability of the first layer, so that most of the detection task can be completed in the first layer.