Research On Cloud-based Image Restoration Algorithm

With the development of the cloud technology, one can access or rent sharedresources of internet conveniently. If huge amount of image resources would beobtained, it may become a prospective problem for the field of digital image thatwhether those resources can be applied to traditional image processing. Due to theconstraints of shooting equipments, natural environment or man-made operations,the original image will be degraded during imagery, transmission or storage. Thedegradation can be alleviated or eliminated by image restoration methods, the resultof some of which may approach the original image with high fidelity. On the basisof variation framework, regularized image restoration thrives in recent years andwins considerable achievements. Whatever image restoration algorithm, all of themshare a common identity that pixels are recovered on the basis of the neighboring orother ones at the same image. If abundant images were collected in a cloud, we mayfind images highly correlated to the original one at high probability and it followsthat we may take advantage of a prior knowledge of this correlation in imagerestoration. In image search system, one can easily get many similar candidates bysubmitting local feature descriptors. Unfortunately, the purpose of image search isnot to recover a high-quality image from its degraded version. In order to integrateimage retrieval methods into restoration and utilize the prior knowledge of therelated content, a cloud-based image restoration algorithm is proposed. The maininnovations of the thesis are as follows:1. Propose a cloud-based image restoration frameworkThe thesis proposes a cloud-based image restoration framework, which appliesrelevant technologies of image retrieval to the procedure of image restoration. Thequery of image search system is a preliminary result restored by an adaptive sparsityrepresentation based regularized image restoration algorithm. After processing, thesearch results can be used to help to do further restoration work.2. Propose an adaptive sparse representation based image restoration algorithmThe thesis analyzes the sparsity representation based image restoration algorithmand proposes an adaptive sparse representation model. In the case of the originalalgorithm, the number of correlated patches becomes more and more with theincreasing of iterations. Adaptively, the threshold of the three dimensionalarrangement in transform domain is set to varying during iterations in our proposedmethod, which removes noises successfully and at the meantime preserves thenonlocal sparsity.3. Propose a cloud-based sparse regularization method The cloud-based sparse regularization is proposed as the core strategy in ourcloud-based image restoration scheme. After feature matching and imagetransformation, a new regularization constraint is built on the similarity relationsbetween the preliminary result and other candidates, which is used to update theobjective optimization function of regularized image restoration.