The Research And Implementation Of Exemplar-Based Image Inpainting Technology Based On Markov Random Field

Based on Cheng algorithm, an modified image inpainting algorithm based on Markov random field is proposed in which the priority function is redefined and the update rule of confidence term is improved. The structure characteristics of unkonown region can be continuously preserved while the texture apperance was well achieved. Another algorithm is discussed which is named the exemplar inpainting algorithm based on multiscale Markov random field model. Instead of the maximum posteriori probability rule used in the single scale, the sequencial maximum posteriori probability rule is used to fuse the inpainting information in the different scale. Image inpainting in the low resolution image was adopted to inform the inpainting process in the next higher resolution image. The multiscale algorithm not only has suppressed some garbage to a certain extent, but also kept the continuity of structure and texture characteristics in the original image.