Image spam detection using frequent item mining technique

Zulkifli, Nor Anis Syazana (2017) Image spam detection using frequent item mining technique. Project Report. Universiti Teknikal Malaysia Melaka, Melaka, Malaysia. (Submitted)

	Text (24 Pages) Image spam detection using frequent item mining technique.pdf - Submitted Version Download (4MB)
	Text (Full text) Image spam detection using frequent item mining technique.pdf - Submitted Version Download (2MB)

Abstract

Spam is usually sent in the form of a text message in which specific words in the message can be used by spam blocking software to prevent the message from reaching our Inbox. With image spamming, the text is placed inside the image in an effort to bypass the spam blocking software. Since images are considered a normal part of a recipient's email message and the spam blocking software is mainly designed for text, the spammer is successful in getting the message to reach our Inbox. One of the previous study has used frequent item mining technique in order to extract features. However, the researcher only considered minimum weightage scheme for feature weight assignment. Thus, the main objective of this project is to generate feature vectors using weightage schemes, which is a maximum weightage scheme assignments. For further investigation, an ensemble method also is applied for weightage schemes. Firstly, sift descriptor is used to represent an image. Sift keypoint make the process of clustering and each of keypoint were collected the cluster number. Bag-of-word feature vectors are generated directly. Then, the frequent items are generated from BOW feature vector using of weightage schemes. SVM classifier is used as image spam classifier to train and produce a models for scheme. Lastly, an ensemble method used for models to obtain the best models. The significant contribution for this project is using the weightage schemes of maximum of the frequent item mining (FIM) technique to generate a feature vectors that is capable of detecting image with better.

Actions (login required)

View Item