An intelligent learning-based watermarking scheme for outsourced biomedical time series data

Abstract

Sharing outsourced data between owners and data mining experts is becoming a challenging issue in biomedical and healthcare fields. Watermarking has been proved as a right-protection mechanism that can provide detectable evidence for the legal ownership of a shared dataset, without compromising its usability. However, the main disadvantage of these conventional techniques is unintelligent, rule-based and they do not directly deal with the data synchronization. Therefore, decoding performance reduces significantly when the watermarked data is transmitted through a real communication channel. This paper proposes an intelligent learning-based watermark scheme for outsourced biomedical time series data. The scheme carries out embedding of watermark data based on modifying mean modulation relationship of approximation coefficients in wavelet domain. In addition, the correlation between modified frequency coefficients and the watermark sequence in wavelet domain is intelligently learnt by a machine learning algorithm. The watermark can be effectively retrieved using this learning algorithm. Experimental results on electroencephalography (EEG) data with support vector data description (SVDD) learning show that the proposed scheme provides good imperceptibility and more robust against various signal processing techniques and common attacks such as random cropping, noise addition, low-pass filtering, and resampling.