Research On Document Image Watermarking Based On Print-Scan Invariants And Double Domain

The documents often need to be shared within the government and enterprise.The shared documents can easily lead to the disclosure of documents in the process of document sharing,and bring significant losses to users.How to quickly locate a leaker based on the leaked document is of great significance for document protection.The document image watermarking can protect the document information by embedding the receiver's logo information and extracting the watermark information to trace the leak source.The embedded watermarked document image need better visual effect.It is very challenging to successfully extract the watermark information after a series of attacks such as scaling and rotation,even in the case of print-scan.Faced with the diversity of transmission methods and attacks,we studies document image watermarking algorithm with anti-printing & scanning,strong robustness and good visual effect.A document watermarking algorithm combines print-scan invariants with double domain is proposed in this thesis.The main research contents are as follows:(1)In this thesis,the print-scan invariants are constructed based on characters.In order to keep a consistent sequence of characters can be segmented before and after print-scan,this thesis design a character segmentation algorithm based on Hidden Markov Model(HMM).In this algorithm,the normalization method of character sample is given to make the size and position of character normalization,so as to reduce the influence of size and position on the eigenvalues.The eigenvalues are constructed by interval mapping using the ratio of the number of black pixels in the sub-images to the number of the black pixels in the character image,and the directional characteristics are extracted.Then the Baum-Welch algorithm with multiple observation sequences is used to train the HMM,so that the recognizer has a better recognition rate.The character is identified and segmented by forward probability and sliding window,and it has good character segmentation effect.(2)The pixel flipping process in the proposed watermarking algorithm is based on Min Wu flipping strategy in this thesis.Because there may not be a pixel block model corresponding to the highest flappable score of Min Wu in the character,and in order to reduce the local distortion and adapt to the character segmentation,we optimize the Min Wu flip scheme.The optimization scheme utilizes a flipping score sorting mechanism to ensure that the pixels with higher scores are preferentially flipped,the four neighborhood is used to prevent the local serious distortion,and whether the row or column of pixels is all white pixels is used to adapt the character segmentation.Because the Min Wu strategy only indicates that the number of black and white connected clusters can be solved by depth-first,no specific algorithm is given,we design a calculation method for the number of black and white connected clusters of a pixel block,and the number of clusters can be accurately calculated.(3)In order to make the watermarking algorithm have anti-printing & scanning,strong robustness and high invisibility,a document watermarking algorithm combines print-scan invariants with double domain is proposed.In this algorithm,print-scan invariants are used to construct eigenvalue matrix to enhance anti-printing & scanning.The eigenvalue matrix is subjected to Discrete Cosine Transform(DCT)to modify the high frequency coefficients to enhance robustness.In order to enhance the invisibility of the watermark,a rule of character screening grouping is designed,and under different conditions,the parameters are used to modify the DCT coefficient to calculate the amount of black pixels to be flipped,and then the optimized Min Wu flipping scheme is used to flip the pixels of the embedded part and the adjusting the part to achieve the watermark embedding.Through the above research on character segmentation,pixel flipping scheme,number of black and white connected clusters of pixel blocks,and document image watermarking,we design a leak source tracking system to capture document images,manage basic information,embed watermark,extract and audit watermark.Experimental results show that the proposed HMM-based character segmentation algorithm has better recognition rate and segmentation effect.Compared with the same type of watermark algorithm,the proposed algorithm based on print-scan invariants and double domain in this thesis has better anti-printing & scanning,has good robustness to the attacks such as scaling,and has a good visual effect.