关键词: 模糊C均值聚类, 单类支持向量机, 隐写分析, 音频质量测度, 音频隐写

Abstract: Concerning the poor adaptability of the traditional audio steganalysis method using two-class classifier to the unknown steganography method, an audio steganalysis method based on Fuzzy C-Means (FCM) clustering and One Class Support Vector Machine (OC-SVM) was proposed. In the process of training, features were extracted from the training audio firstly, including the statistical features of the spectrum of the Short-Time Fourier Transform (STFT), and the features based on audio quality measures; and then FCM clustering was executed on the extracted features to obtain C clusters; finally the extracted features were trained by the OC-SVM classifier with multiple hyperspheres. In the process of detecting, the features were extracted from the testing audio, and the testing audio was detected according to the boundary of the OC-SVM with multiple hyperspheres. The experimental results reveal that,for some typical methods of audio steganography, this steganalysis method can detect accurately, when the embedding capacity is full, the total detection accuracy is 85.1%; furthermore, compared with the method of K-means clustering, this method can improve the detection accuracy by at least 2%. This steganalysis method is more universal than the steganalysis method using two-class classifier, and it is more suitable for the detection of the stego-audio whose steganography method is unknown beforehand.

Key words: fuzzy C-means clustering, One Class Support Vector Machine (OC-SVM), steganalysis, audio quality measure, audio steganography