Image Reconstruction on the Basis of a Textural Geometrical Model

摘要

A new method for reconstructing two-dimensional signals on the basis of a textural geometrical model and separate restoration of the structure, texture, and uniform image areas is presented. An approach that provided the best reconstruction result is used for each class of the areas. In the first stage, the Laws characteristics for extracting the textural features and clustering the images are obtained. Then, the spline inter-polation is applied to restore the contours of the boundaries of objects. Furthermore, the texture is restored by exemplar-based method, which consists in searching for the similar blocks on the initial image and copying them to the area with absent pixels. To restore the uniform areas, the Telea method is used. Here, it is suggested to calculate the perceptual hash in searching for the resembling blocks, which allows one to decrease the computing complexity. Research on a new approach is presented by several examples showing the efficiency of the algorithm at removal of large and small objects on test images.