Study On Compressed Sensing And Digital Watermark Preprocessing

With the rapid development of information security, people pay more attention to the copyright of multimedia products. Digital watermarking can insert some useful information into the original data, by some kind of embedding algorithm without any effects on the carrier. If needed, the embedding information can be extracted by corresponding algorithm to prove the truth and integrity of the original data and protect the benefits of copyright owner. At present, most studies about digital watermarking focus on the embedding and extraction algorithm, and ignore the importance of watermark preprocessing. Compressed Sensing (CS) theory combines signal sampling and compression, realizing the collection of signal and encryption at the same time. CS has become the research hotspot in the field of information security.This thesis combines CS and digital watermarking technology according to the characteristics of watermark image, mainly studies the structure of CS measurement matrix and preprocessing of watermark image. It analyzes the existing watermark preprocessing algorithms. This thesis divides watermark preprocessing algorithm into four categories, respectively, based on pixel scrambling, encryption algorithm, coding and the hybrid algorithm. And then it proves that it is feasible to use pseudo-random sequence to generate CS measurement matrix. This thesis uses m and Gold sequence to generate CS measurement matrix and optimize them with QR decomposition, then uses them to compress and reconstruct one-dimensional signals and two-dimensional images. The experiment results prove that measurement matrix can reconstruct original signals well.At last, this thesis introduces three different kinds of watermark preprocessing scheme. Researches are mainly concentrated on schemes based on Arnold transform, CS and m sequence. It verifies the feasibility of the three schemes by experiments, and compares the ability of resisting JPEG compression, salt and pepper noise, median filtering and shear attack.