Conference Paper

A Discriminative Classifier Learning Approach to Image Modeling and Spam Image Identification.

Conference: CEAS 2007 - The Fourth Conference on Email and Anti-Spam, 2-3 August 2007, Mountain View, California, USA
Source: DBLP
  • [Show abstract] [Hide abstract]
    ABSTRACT: 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%.
    Journal- Chinese Institute of Engineers 05/2014; 37(4). DOI:10.1080/02533839.2013.815005 · 0.21 Impact Factor
  • Source
    [Show abstract] [Hide abstract]
    ABSTRACT: Email spam is a major problem for the sustainability of the Internet and global commerce. Every day million of emails are sent across by spammers to targeted population to advertise products, services, spread dangerous software etc. Currently a number of spam detection algorithms have been proposed in the literature to classify email spam. Most of these algorithms can be categorized as meta-data based, content based, behaviour based, etc. The existing literature has heavily focussed on advancing one or the other kind of approach i.e. there are a lot of algorithms doing content based or behaviour based detection, however there is not much work done that evaluates the effect of applying different algorithms in a step by step and/or iterative manner to achieve optimised classification. This research would aim to evaluate existing algorithms and propose a spam detection framework to automatically choose the correct algorithm sequence to do the classification based upon intelligently identified heuristics from the email profile.
  • [Show abstract] [Hide abstract]
    ABSTRACT: Image spam continues to be one of cyber security problem today. Spammers used image spam as a technique to by-pass conventional email filters. Anti-Spammers used image classification as a method to detect images spam by extracting different features of the image. One of the important features used is color features. Several works used different color analysis to differentiate image spam, most of these works used supervised methods trying to differentiate computer generated images which is mostly like to be a spam and natural images. Supervised methods have its weaknesses, such as high cost in computation, requires training data, and rapid changes in spammers behaviors. This paper develops an unsupervised method using HSL geometric model (Hue, Saturation, and Luminance) to distinguish computer generated (CG) and natural images. Rules and Heuristics are defined by using HSL variables. The proposed method mainly depends on Saturation and Lightness values and their histograms. Experiment results shows that the combination of these variables can give high classification accuracy results.
    Advanced Computer Science Applications and Technologies (ACSAT), 2012 International Conference on; 01/2012

Full-text (2 Sources)

Available from
May 23, 2014