Malicious Image Detection Using Convolutional Neural Network

Images have become a threat to the security of the systems and networks since JPEG headers are concealed with malicious payloads. Header in a JPEG has many segments which can be manipulated with executable codes to prepare for malware attack. Images are usually perceived as harmless and non-risky by the users so they have become the focus of attention for carrying the cyber-attacks. Security threats in systems and networks which are caused by malicious images, are needed to be minimized by introducing a detection technique, a technique which can involve features of headers. In our proposed method JPEG headers are transformed into grayscale images to employ classification. Convolutional Neural Network based model is proposed which aims the detection of malicious images. We have used a dataset of JPEGs which was collected from different honeypots installed by CRC of Bahria University. Dataset contains 1100 malicious and 1100 benign images to employ the detection method based on deep learning. We have achieved 96% accuracy. Our method of malicious image detection would help everyone to prevent the malware attacks which are carried through images.