A Study On A Document Watermarking Using Background Texture For Text Images

Recently, information leakage has become one of the serious problems in thenetworked society. Surveys reported that about50%of information leakage problemsare due to inadequate treatment of paper-based media (printed documents).Information security problems have become prominent, information security in theinternet environment is under a lot of new challenges.The paper put forward a document watermark using background texture for textimages. Firstly we introduce the basic concept of the digital watermarking technology.Then we do a basic study on a document watermarking using two kinds ofblock-patterns for text images. In this basic study we generate background picturewith halftone square patterns successfully.On further research, we proposed a document watermarking method that usesnew background texture patterns. In our method, we use4kinds of block pattern torepresent watermark information as well as the marker information, where each blockpattern quantizes the gray-scale background texture into5levels. Thus, block patternsalso represents gray-level of the background picture. Marker information achievesprecise positioning and synchronization for watermark fields, which makes thismethod resistant to cropping attack.Then we use Gabor filter for efficient and stable detection of marker andwatermark information. We convolute two kinds of Gabor filters whose phases are0and π/2to specifying marker information and another two kinds of Gabor filterswhose phases are π/4and3π/4to specifying watermark information. We analysetexture to determine a bit of a block-pattern by the K-means algorithm and at last wecreate a linear discriminant function based on neuron model to conclude a singlewatermark bit.Due to the marker information for synchronization, we could robustly detectwatermark information even from the partially cropped document. From the results, when only25%of the document is utilized, perfect detection of the watermarkinformation can be achieved for m≧32.Also, we could achieve high watermarkdetection rate against copying. We can detect almost100%of the water informationwhen the number of copying is3and for m≥32. The experimental results show thehigh application value of this watermark.