Abstract:Traditional spam filters are facing more and more challenges with the rapid growth of image-based spam. Previous works have leveraged OCR techniques and text classifiers for image spam detection, which are time consuming and CPU intensive. In addition, OCR can be easily tricked by noise-and contentobscuring elements added by spammers. In this paper, we propose a novel approach aimed at detecting the "amount" and the "type" of noise due to the use of those techniques against OCR tools. Firstly, we propose a specific method for text region localization using steerable filter and morphological processing, which separates images into text-regions for OCR content extraction and background-regions for noise analysis. Next, wavelet transform is used for constructing noise feature image of the background-region, based on which noise measurement and classification can be completed. Experimental results show that our method can locate the text region accurately, and the results of noise analysis can effectively reflect the noise interference of spam images, which can be viewed as complementary to the approaches based on OCR tools for further reducing false positives of the image spam filters. © 2012 Binary Information Press.

Research of Feature Extraction and Selection for Spam Image Identification

A Survey for Study of Feature Selection Algorithms

Feature Selection for Image Spam Classification

An Effective Method for Distinguishing Photograph and Graphics

Text Region Extraction in Image-Based Spam Email

Image Spam Identifying Algorithm Based on Color and Corner Feature

Graph-Based Semi-supervised Feature Selection with Application to Automatic Spam Image Identification

Method of Image-Based Spam Filtering of E-Mails Based on Image Similarity Detection

Research on Spam-Filtering System Based on Visual Features Analysis

Image Spam Classification Model Based on Color and Edge Histogram Statistics

Research on Image Retrieval Method Based on Color Classification

Spam image discrimination using support vector machine based on higher-order local autocorrelation feature extraction

Text Localization In Spam Image Using Edge Features

Feature Importance Analysis for Spammer Detection in Sina Weibo

Image Spam Filtering Based on Gradient and Color Feature

Efficient Modeling of Spam Images

Image Spam Classification Based on Low-Level Image Features

Fusion of text and image features: A new approach to image spam filtering

A Survey on Key Technologies of Image Spam Filtering

Noise analysis for text-based spam images

Image Spam Identification Method Based on Gray-Gradient Co-Occurrence Matrix