Content-Based SPAM Image Detecting Technique Research

With rapidly growing, spam image has become the mostly type of SPAM currently. How to deal with spam image has been to a global problem and none of us can evade. Up to now, there is not so much working on Anti-SPAM image research. And most of work we do now is just a kind of explore, and there is a lack of corpus library in public. With a start from a technique point of view, on the base of system studying and summarizing the newest anti-SPAM production, this research went deep into SPAM image detect technique, and present a new method in SPAM image detection, and improve one kind of arithmetic of extracting text area from image. It has been proved true in SPAM image detection after experiment on that arithmetic.Paper includes the writer's main work in aspects below:1. Introducing concept of SPAM image along with its harm and both domestic and overseas status of research in this field;2. It spends me 6 months establishing a SPAM image corpus via two e-mail accounts; 3. Summarize SPAM image detection technique, including color feature based detection technique, figure feature based detection technique, texture feature based detection technique and object based detection technique;4. Present a new image detection arithmetic: framework detection arithmetic. By extracting and matching the framework feature of image, SPAM image can be detected.5. Improve a text area extraction arithmetic. Firstly, wipe off the noise by wavelet technique, and than using corner detection to extract text area of image, finally estimate if the image is a SPAM or not by the proportion of text area. It has been proved by experiment result that the improved arithmetic is better in robust to noise and higher in effect.