| Image denoising,working as a fundamental restoration problem,takes an important role in the whole image processing.It is worthy studying and discussing due to its importance.There has been a growing interest in image denoising algorithms based on overcomplete dictionary in the past decade.The K-SVD algorithm is famous for its efficiency in building adaptive overcomplete dictionary.With the development of image denosing algorithms,a series of technique methods can still be used to boost image denosing algorithms.This thesis discusses the K-SVD algorithm and SOS boosting algorithm.The shortcomings of of the two algorithms are stated in thesis.On the basis of those discussion,some contributions are made,which include:1.The principle of the K-SVD dictionary algorithm and the reasons for limiting the performance of the K-SVD algorithm are stated in the thesis.Then a new K-SVD algorithm based on regularized model which is extended to the AK-SVD algorithm is proposed.Simulation results show that the K-SVD algorithm based on the regularization model and the dictionary generated by the AK-SVD algorithm have higher training accuracy.The time complexity of the algorithm can be decreased by the proposed algorithm.2.The efficiency and convergence of the SOS boosting algorithm are discussed.And a new termination criterion based on difference PSNR is introduced into the SOS boosting algorithm to improve the adaptive performance of the original algorithm.Simulation resulst show that the improved SOS framework can exit the iteration in time,which saves a lot of time.3.The proposed K-SVD algorithm based on regularized model and the improved SOS boosting algorithm are combined together to improve the performance of image denoisng.Also,some unique features are found in the context of K-SVD and SOS boosting algorithm.Simulation results show that the SOS denoising boosting algorithm based on the proposed K-SVD algorithm can significantly improve the denoising performance of the original algorithm. |
PDF Full Text Request