Research On Adaptived Patch Matching Ramdom Search Image Inpainting

As one of the main carriers of Internet information,image information can be transmitted to the receiver more accurately and directly,which is affecting the way of live and work of human being now.Due to the wide application of digital image processing technology,more and more researchers have been involved in the research work of this field.Image inpainting technique is also called image completion which is the branch in the field of digital image processing and computer vision.It has always been a research hotspot and its main idea are:First develop a good inpainted strategy,then utilizing known information in the original image based on the above strategies to inpaint the damaged areas to achieve the purpose of inpainting the entire original damaged image.At present,there are a lot of image completion altorithms.They can be divided into two broad categories:inpainted algorithm based on partial differential equation and inpainted algorithm based on texture synthesis.These two methods have their own advantages.The former is mainly aimed at smaller images in damaged areas and the latter was contrary to the former.This paper mainly focuses on image completion technology.First of all,it states the basic principles of the technology and then analyzes and summarizes the current research methods all around the world.Secondly it will introduce the basic knowledge of image inpainting.After that,we will reorganize?analyze and summarize several typical image inpainting algorithms.On the basis of the above methods,it proposes two kinds of image completion methods in this paper and the specific contents are as follows:Firstly,this paper presents a Multi-scale Decomposition Based k-nearest-neighbor-field Random Search for Image completion.The damaged image is decomposed by multi-scale based on bilateral filtering down sampling and using similarity between patches as a measure method.At the same time,it presents a k nearest neighbor field search algorithm based on minimum heap to obtain the best matching patch.Then,utilizing the robust priority function to determine the next damaged patch that shoud be inpainted.Repeat the operation until the layer is inpainted.The lower coarse layer is reconstructed using the bilateral filtering based up sampling and the information of current image layer is inpained,so as to get the final result with iterative completion.Finally,the simulation results verify the effectiveness of the proposed algorithm.Secondly,on the basis of the former method,this paper proposes an image completion using adaptive matched patch based narrowband optimization.The basic idea of this method is to use the color and structure information of the damaged patch of each layer to calculate the size and the number of matching patch.Using a sum of squares of difference to achieve patches of corresponding scale.Then,to achieve the matched patch sequence and optimized based on narrowband model to inpaint damaged boundary according to minimum heap k nearest neighbor field random search in that patch.The damaged boundary was inpainted from outside to inside until the damaged area was completed.