Research On Digital Watermarking Algorithm Based On Texture Directionality In Non-subsampled Shearlet Domain

Singular characteristics in images or higher-dimensional signals can effectively be represented by multiscale geometric analysis. Shearlet is one of the latest multiscale analysis functions. It is more sensitive in capturing direction features and has strong sparse representing ability. In this thesis, on the basis of deeper research of shearlets , non-subsampled shearlet transform is applied to the digital image watermarking field. Non-subsampled shearlet transform discards subsampling process of shearlet transform and overcomes the shortcomings of spectral aliasing phenomenon. In addition, it has good localiztion, multi-direction and optimal sparse representation characteristics. First,the image is decomposed into a plurality of directional subbands by non-subsampled shearlet transform. Then the watermarking embedding position is determined according to the the values of texture features. Finally, two kinds of watermarking algorithms in Shearlet domain are proposed by using bidiagonal singular value decomposition and principal component analyses. The main work is as follows:(1) Selection of watermarking embedding location. After images are transformed into frequency domain,there are many directional sub-bands. Choosing a appropriate sub-band as the embedding position for the watermark plays an important role in the performance of the watermarking algorithm. This thesis analyzed the image texture features and chose the texture directionality as criterion to determine which sub-band is more suitable for watermark embedding. Texture features contain high frequency information of the image. These feature information can be embedded with watermark. The reason why do so is that it complies with the human visual characteristics and also improves the robust performance of watermarking.(2) A non-blind watermarking algorithm based on bidiagonal singular value decomposition and non-subsampled shearlet transform is proposed. After strongest texture directional subband is determined, the watermark is embedded by adding them to bidiagonal singular values of the subband. Bidiagonal singular value decomposition not only has the stability of general singular value decomposition and other excellent features, but it also brings little distortion to the original images. By doing so, higher PSNR value can be ensured. At the same time,safety of the watermark can also be improved because of its unique calculation process.The experimental results show that the embedding of the watermark does not cause much damage to the original image. And the watermark can be extracted completely after many kinds of attacks.(3) This thesis proposes a blind watermarking algorithm, combining with image texture directionality and principal component analysis. After the watermarking embedding position was determined, the principal component analysis is used to obtain the dimensionality reduction information, which can optimally represent the main information of the image. The embedding and blind extraction of watermarking are realized by using the adjacent pixel average value. Dimensionality information contains much energy of the image. So the watermark is able to resist many forms of attacks, thus leading to good robust performance.