Study On Zero-watermarking Algorithm Based On Non-subsampled Shearlet Transform

A zero-watermarking algorithm was proposed and discussed in this thesis,which combines inherent image directionality features with non-subsampled Shearlet transform.In order to extract the coefficients that have close relationship with image directionality features,k-menas clustering method was applied to image direction sub-band in the Shearlet frequency domain.According to the former studies,directionality features like edge or texture are everywhere in a given image.Shearlets' direction sensitivity and anisotropy not only make it a great tool for image sparse representation,but it can also accurately capture the inherent image directional features.What the thesis had discussed are listed below:(1)A zero-watermarking algorithm based on Shearlets' high directionality features capturing ability was proposed in this thesis.In the Shearlet frequency domain,the inherent direction feature information is included in the coefficients of different direction sub-bands.Direction information intensity,sum of pixels of single sub-band in the time domain,was proposed to judge which sub-band is suited for watermark embedding.After the embedding location is determined,the sub-band is partitioned into several parts,the number of which is the same as actual watermark.Then,the matrix 2-norm of each part is calculated,followed by binaryzation operation.The embedding process is accomplised after zero-watermark is exclusive or(XOR)actual watermark.Experimental results reveal that the proposed zero-watermarking algorithm has outstanding robust performance,especially in anti-rotating attacking when compared with the algorithms based on DWT.(2)K-means clustering method is applied in the sub-band selected to filter the coefficients.The reason why do so is to partion the coefficients that have close relationship with the image directionality features.Then these coefficients are selected to construct zero-watermark.Thus the direction sensitivity of non-subsampled Shearlet is utilized as much as possible.At the same time,by doing so,we actually take the advantages of image directionality features.The experimental results demonstrate that the k-means clustering participated algorithm has elevated anti-attacking ability to a higher level.