Malware variants detection based on opcode image recognition in small training set
Tingting Wang, Ning Xu · 2017
This paper proposes a method to solve the problem that the accuracy of malware detection is severely limited within the small training set. Firstly, we disassemble the malware binaries into opcode sequences and convert the opcode sequences into images. Then, we enhance the images by using histogram normalization, dilation, erosion, and extract feature by using principal component analysis (PCA). Finally, training and detecting by using classifier SVM based on RBF kernel function. The experimental results show that the image recognition can improve the accuracy of malware detection; in case of small training set, the detection accuracy is improved and the detection time is much less than the time of KNN.