Feature Selection Based on Neighborhood Entropy for Data Hiding Analysis
Yanqi Xie, Yumin Chen · 2019
The feature extraction process in data hiding analysis produces a large number of high-dimensional, redundant and uncertain data, resulting in high computational overload and low recognition rate of data hiding analysis systems. Therefore, this paper proposes a feature selection method for data hiding analysis based on neighborhood granulation and neighborhood entropy, exploiting the theory of neighborhood granular computing and the uncertainty measurement method with information entropy. First, the neighborhood granular computing theory is used to granulate the features of data hiding analysis systems. Second, the information entropy measurement of neighborhood granules is defined, and the feature importance is constructed by neighborhood entropy to evaluate the importance of features in data hiding analysis. Furthermore, a feature selection algorithm is designed to select a few features with strong discriminative characteristic. Finally, an SVM classifier is trained using the selected features under different algorithm configurations including one data hiding method, two feature extraction methods and different information hiding amounts. Experimental results show that the proposed method can obtain compact features for data hiding analysis and improve the performance of the classifier on these features.