Image restoration using piecewise iterative curve fitting and texture synthesis

In this paper, we present a novel scheme for parsing images into medium level vision representation: contour curve and region texture. This scheme is integrated with piecewise iterative curve fitting and texture synthesis, in which an original image is analyzed at the representation side so as to obtain contours and local texture exemplars. The contour of each region will be processed piecewise by iterative curve-fitting procedure. At the reconstruction side, transmitted coefficients denoted to structural information of regions will be used to reconstruct contours. Texture synthesis algorithm is adopted to restore the rest. Experimental results indicate that our scheme can be done successfully for images with massive structural and stochastic textures. Compared with JPEG, our method can get more coding gain but remain vast information people care.