Dunhuang Frescos Segmentation Based On Markov Random Field And Graph Cuts

Because of some natrual and man-made reason,Dunhuang Fresco has many disrepair situation,such as broke,blur,fading,dark,displacement and so on.Hence,it necessarily need to be inpainted.As an important technique of computer vision,image segmentation achieves good inpainting result of Dunhuang fresco,such as color fading and information losing.Experiment shows that it could enhances accuracy of inpainting if we segment some disrepaired region first.This paper focus on the image segmentation method based on the framework of Markov Random Field(MRF),and modeling algorithm under the image feature of maximum a posteriori(MAP).First of all,a popular optimal task named Graph Cut is researched by the paper and a new interaction image segmentation method which based on MRF and Graph Cuts is proposed.By using a group of filters, this method extracts low level features from images and obtains feature vector.At the beginning, some feature with high distinguishment effect is choosen and setted as the measurement rule of clustering.After that,seeds point respectively belongs to foreground and background are obtained by interactive way.Then,statistic model of foreground and background is obtaind by using clustering method and image features of seeds.Through setting clustering number as tag of image,the similarity between pixels and seeds are measured by Graph Cuts.Meanwhile,by minimizing global energy,the algorithm obtains convergent result of image segmentation.Experiment proves that this comprehensive algorithm achives effective results in the field of Dunhuang fresco segmentation.In the next place,the algorithm which introduced upon involves clustering vector made by image filters.So,this paper applies a concept of non-correlation to reduce the errors brought by normal clustering process.The function of non-correlation includes neighbourhood information of pixels and adds different weight to neighbourhood of pixels,in order to make the clustering result more accurate.Once again,we import Watershed Algorithm and apply two steps to reduce calculated amount.Firstly, making initial segmentation by using watershed algorithm.Secondly,take the hole areas replace pixels to calculate the algorithm again and enhance the execution efficiency of watershed algorithm.At last, this paper applies the methods to segment disrepair area of fresco and using colorazation to inpaint these areas.As a new thought of Dunhuang fresco inpainting, this paper applies Welsh colorazation algorithm to paint disrepair area.