Text Image Watermarking Resisting To Print-scan Attack

The development of electronic commerce makes text image widely used.Print-scan is an important way of text images transmission,and it is also a common way of text images information leakage and theft.The text image watermarking technology hides the information representing copyright in the text image.When there is a copyright conflict,the copyright is verified by extracting the watermark,which provides an effective solution for text image copyright authentication.Watermarking algorithms are usually measured by invisibility and robustness,which restrict each other.An excellent watermarking algorithm should strike a balance between robustness and invisibility.While ensuring the invisibility of the watermark,try to improve the robustness of the watermark.The text image watermarking methods resisting print-scan attack can be divided into transform domain method,spatial domain method and dual domain method.In the transform domain method,the text should be processed as binary images with only black and white after embedding the watermark.This method is not suitable for binary text images because the watermark is easily lost during the binarization process.The spatial domain method directly modifies the text pixels to embed the watermark,which makes the watermark less robust.The dual domain method combines transform domain and spatial domain methods to embed watermarks,which not only can be used for binary text images,but has good robustness.The current dual domain method mostly adopts the method of embedding a watermark once,and the bit error rate is high when extracting the watermark.Secondly,the pixel flipping strategy is an important technique to balance the invisibility and robustness of the watermark in the spatial domain method,which is divided into single-pixel flipping strategy and single-contour flipping strategy.The singlepixel flipping strategy is easy to cause human eye perception after flipping the pixels,and the single-contour flipping strategy is easy to cause the wrong watermark to be embedded.The above two points make the watermarking robustness of the dual domain method need to be improved.In view of the above analysis,this paper proposes a new dual domain text image watermarking algorithm,which can improve the resistance of the watermark to print-scan attack on the premise of ensuring that the watermark is visually invisible.The innovative work of this paper is as follows:(1)A watermark preprocessing compression method is proposed.The watermark image is compressed by white block coding before embedding,so that the watermark is dynamically embedded multiple times under the condition that the total embedding capacity is certain.This method can reduce the bit error rate of single embedded watermark and improve the robustness of watermark.(2)A multi-contour pixel flipping strategy is proposed.All contours in the connected components(areas where all pixels are connected)in the text are regarded as modifiable objects.Each contour is modified one by one,and the pixels that can be flipped are selected according to the difference between the old and new contours.This method has high invisibility because embedding the watermark according to the modified contour,and at the same time reduces the possibility of embedding the wrong watermark in the single-contour flipping strategy,further improving the robustness of the watermark.(3)An optimized dual domain text image watermark embedding and extraction algorithm is proposed.Embedding watermarks by print-scan invariants of connected components of text images: the discrete cosine transform coefficients of the feature matrix constructed by print-scan invariants are modified according to the watermark information,and print-scan invariants of embedded watermarks is obtained by inverse transformation;then pixels of the connected component are modified by using the multi-contour flipping strategy,so that the print-scan invariant is changed into the print-scan invariant embedded watermark,so as to achieve the purpose of embedding watermark in the text image.The algorithm is designed based on the constant quantity of text images during the printing and scanning process,and combines the discrete cosine transform and multi-contour pixel flipping strategy to embed the watermark to improve the resistance of the watermark to the print-scan attack.The experimental results show that the watermarks' invisibility of the proposed algorithm increases by 8.81%,5.16% and 0.03% respectively on the basis of PSI+SPFS,PSI+SPFS+DCT and PSI+SCFS,and the robustness increases by 49.82%,17.3% and 23.31% respectively.The proposed algorithm can effectively improve the watermark's resistance to print-scan attack under the premise that it is not visible.In addition,the watermark can resist rotation,compression,noise and other attacks within a reasonable range.