Research Of Image Editing Based On Patch Table

With the popularity of smart phones and digital cameras,image as a carrier of information,is playing an increasingly important role in our life.The increase of the number of images makes image editing gets more attention from researchers.Image editing encompasses the processes of altering images such as image enhancement and restoration,color transform and artistic style.In many image editing applications,the approximate nearest neighbor search is a vital important part.In this paper,we propose an efficient patch query algorithm-patch table based on isometric mapping.The feature of patch will be maintained after using isometric mapping to reduce the dimensionality,and this will help improve the accuracy of the process of patch match.In the process of constructing patch table we firstly populate the table with patches from image dataset,then fill holes in the table with Voronoi transform,finally compute k-coherence.Then,for the problem of image color transform,we propose a novel method for image color transform based on image attribute.The achievement of image color transform also can be separated into three parts.The first operation is finding match image sharing the similar scene for input image,then searching the target image with the expected attribute that needs to transform,finally computing the local transform between match image and target image,and applying the local transform to input image to get the final result.Finally we present a novel method for artistic stylization for face.The achievement of artistic stylization for face can be separated into three parts.The first operation is using grabcut algorithm to detect and segment face area from the input image,the second step is synthesizing the artistic image for face,the last step is assembling the artistic image to original image.The experiments we did show that our methods have a good result and advantage over other methods especially in time complexity.