| Summary: | 碩士 === 國立臺中技術學院 === 資訊科技與應用研究所 === 97 === In this thesis, a study on digital watermarking technology by applying support vector machine (SVM) for relational database integrity authentication is proposed. Owing to the elegant machine learning ability of SVM, SVM is used to learn and predict the correlation for relational database, and then to perform the purpose of the database content integrity.
In Chapter 3, an effective solution based on the fragile watermarking technique is proposed by exploiting the trained SVR predicting function to distribute the digital watermark over the particular numeric attributes. While the watermark bit is equal to 1, add 1 to the predicted value and replace the original attribute value with the new predicted value. Otherwise, while the watermark bit is equal to 0, corresponding original attribute value is replaced by the value which is subtracted 1 from the predicted value. In detection phase, the same SVR predicting function is used to generate predicted value, and if the absolute difference value between predicted value and attribute value is more than the designed fixed value, like one, then the database content is determined to be tampered with.
In Chapter 4, the proposed watermarking scheme based on SVR prediction, which exploits the digital watermarking technology for guaranteeing the database integrity underlying distortion free of database content. The proposed scheme employs SVR predictive function to obtain characteristic of the database and uses Huffman coding to encode the characteristic for compressing important payload information. In detection procedure, minor and necessary additional payload information of the database is used to accomplish tampering detection.
Eventually, Chapter 5 proposed a reversible fragile watermarking based on SVR prediction for authenticating database integrity with original values recovery. While the protected database is modified by malicious users, the trained SVR predicting function is used to generate difference and extract embedded watermark bits to detect modified tuples. Furthermore, the proposed scheme is capable of recovering any original value after a tamper-free recovery procedure where the embedded watermark bits are properly extracted. In other words, the proposed method really has the power of effective detection and locating malicious tampering, achieving database authentication and recovering content integrity.
|
|---|
