Research On Automatic Watermarking Removal Based On Clustering Algorithm And Fast Marching Method

With the improvement of people's copyright awareness and the widespread network sharing of information,digital watermarking technology has been widely used to protect the prop-erty rights of authors.However,many businesses abuse digital watermarking technology by adding their own advertisements or blocking key parts in some public electronic materi-als,which greatly affects the visual experience of readers.In this paper,image processing techniques are studied to remove watermarks through some algorithms.This paper first describes the reasons for the generation of digital watermarks,the back-ground of their widespread use,and differentiation method based on generation mode.Then,the typical algorithms and models proposed by previous research in recent years are intro-duced from three aspects:image segmentation,image inpainting and watermark removal.In addition,their contribution,advantages and disadvantages of these algorithms are analyzed for the part with greater reference value.After that,the paper analyses several typical image segmentation algorithms and their math-ematical models,and finally chooses to use K-means clustering algorithm to segment the watermark part of the image.The algorithm is efficient and easy to implement,and it has excellent results in color-based image segmentation.This paper provides solutions to the problems that arise in the running process of the original algorithm and improves them through experiments:(1)Transform the input picture from RGB color space to Lab color space which emphasizes the color change of the pixel points to suit the processing environ-ment of K-means clustering algorithm in two-dimensional coordinates.(2)Introduce canopy algorithm,preprocess the input data before using K-means clustering algorithm,and roughly predict the number of clusters.(3)Divide the sample into two sets and use the point farthest from one of them as the cluster centroid to preset the initial location of the cluster centroid.(4)Introduce cost function and damping factor to slow down the convergence speed of the algorithm appropriately and reduce the probability of oscillation by suppressing the ampli-tude of movement of cluster centroids.(5)Mark possible outliers and reduce the Euclidean distance from the cluster centroid during the algorithm operation,thus reducing the impact of outliers on cluster partitioning.Next,the paper compares several widely used image inpainting algorithms and chooses fast marching method to repair the area where the watermark is located.On the premise of fast running,this algorithm introduces a gradient function to estimate the texture information of the image,which has a satisfactory effect in repairing images with small area.On this basis,the research puts forward the improvement ideas for the existing problems:(1)Enlarge the neighborhood area of each point to be repaired.Therefore,the gradient direction for refer-ence is increased,and more texture information is provided for the repair of sample points.(2)Increase the number of reference sample points and classify them by gradient direction to improve prediction accuracy.(3)Extract the texture factor from the weight function,and select all sample points in the direction of maximum gradient amplitude to participate in the calculation.In this way,the weight function is simplified and the proportion of texture direction is increased.(4)Increase the arrival time of the sample points with large gradient direction to make the algorithm gives priority to repair the outer area of the image where the pixel value changes more smoothly,so as to reduce the adverse impact of error information on the internal points to be repaired.The experimental results show that the method described in this paper can realize the auto-matic removal of image watermarks,which reduces the mean square error while retaining the advantage of relatively less time spent in the original algorithm.