Research On Image Non-local Means Denoising Algorithm Based On Probabilistic Feature Descriptor

Images denoising is a basic subject research in the field of image processing and computer vision.Its purpose is to remove noise while preserving image details as much as possible.So,image denoising is a hot topic in the field of digital image processing,which can improve the image quality.The existing image denoising methods can be divided into local methods and nonlocal methods.Nonlocal means(NLM)filter algorithm which takes full use of the self-similarity of images is currently a image processing method with strong ability of denoising.However,NLM algorithm also has some shortcomings,especially the calculation cost is relatively large.Based on this deficiency,we in-depth study a new local binary descriptor(LBD),the robustness is improved by binary description of the local structural features of the image,and the similarity between pixel blocks is measured by logical operation instead of Euclidean distance,the application of this descriptor can significantly improve the operation speed.However,NLM algorithm based on LBD descriptor(NLM-LBD)is not suitable for multiplicative speckle noise,which is common in medical imaging(eg,OCT image),therefore,we have deeply studied an improved non-local mean algorithm based on probability(PNLM),mainly in the denoising effect on the classic NLM algorithm to make improvements,although the improved algorithm based on the probability of multiplicative speckle noise removal advantage is obvious,but the disadvantage in the operation speed is obvious.Based on the lack of real-time,use LBD descriptor,a non-local mean denoising algorithm based on probability descriptors(PNLM-LBD)is proposed by combining the LBD descriptor and the improved algorithm based on probability,the pixel block is binary-coded at the same time as the probability is calculated,this algorithm achieves the combination of denoising effect and real-time property.Finally,the improved algorithm is parallelized with GPU.The experimental results show that the NLM-LBD algorithm is effective for denoising the additive Gaussian noise and the real-time performance is very good;PNLM algorithm for the multiplicative speckle noise removal is pretty obvious;PNLM-LBD algorithm to OCT image multiplicative speckle noise removal achieved good results and real-time has also been improved;The parallel computation of PNLM-LBD algorithm using GPU can make the computing speed about 45 times higher than CPU,which can meet the requirement of medical image real-time processing.