Research On Digital Watermarking Algorithm Based On Sparse Transform

With the rapid development of multimedia technology,digital products have been more and more widely used.However,copyright protection and information integrity protection of digital products in the open environment have become an important issue to be solved.As an effective method to realize multimedia copyright protection,digital watermarking technology has attracted extensive attention from researchers.Digital watermark including audio watermarks,image watermarks,and three-dimensional model watermarks.This article takes image watermarking as the research object,focusing on the feasibility and application of sparse transform in digital watermarking.The main research work is as follows:(1)The compressive sensing based digital watermarking algorithm solves the sparse coefficients by using the random measurement matrix at the detection port,but such operation usually results in time-consuming detection process,lacks of practicality and the quality of the watermark reconstructed by compressive sensing is generally low.To address these issues,a novel watermarking algorithms using the sparse transform and Laplacian pyramid is proposed.Our algorithm adaptively selects the frequency domain coefficients by training a sparsity domain with a low sparsity.Then select the minimum value of the global coefficients in the sparse domain for embedding.At the same time,Laplacian pyramid is used to enhance the transparency of the watermarking algorithm.We only embed information in low-frequency subband images,and high-frequency subband images are used to preserve and restore image details.As a result,when detecting the watermark,one can adopt the matrix multiplication strategy directly to transform the low-frequency subband of the watermarked images into the sparse domain for detection,which can avoid the process of re-calculating the sparse coefficients.Extensive experiments demonstrate that our proposed method can costs less time for detection,has better transparency,and moreover is robust against attacks.(2)In order to extract the watermark from the sparse domain without original image,a blind watermarking algorithm based on sparse transform and Alpha-stable distribution is proposed.We segments the carrier image uniformly.Then,the segmented image patches are expanded into vectors to learn the sparse transform.Then we use the Alpha-stable distribution to model and analyze the components of the sparse domain.Sparse domain components with similar distributions are selected to construct an embedded vector.Then,we combine the idea of spread transform dither modulation to embed watermark information bits in this vector.And each embedding vector of the image block uniquely corresponds to one watermark information bit.When extracting,the watermark can be detected in sequence without the need of original image,only by transforming the host image to the sparse domain.