Research On OCT Image Denoising And Segmentation Based On Similar Low Rank Prior And Curve Evolution

Optical Coherence Tomography(OCT)is an imaging technique by measuring the reflected light intensity of the object.Since it was put forward in 1991,it has been paid lots of attention due to its unique advantages.At present,OCT imaging technology has been widely used in ophthalmic retina.OCT imaging technology is based on the interference of light,so OCT images inevitably are contaminated by speckle noise caused by optical coherence.Speckle noise seriously damages the visual quality and recognition of OCT image,and then interferences clinical diagnosis.Therefore,OCT image despeckling is of great significance for the development and clinical application of OCT imaging technology.Common ophthalmic diseases,such as glaucoma,vitreous membrane warts,macular edema and other fundus diseases,are closely related to the morphological structure of retinaSpecific ophthalmic lesions which present as plaques usually appear on the corresponding retinal layer in OCT retinal images.Atpresent,in clinical diagnosis the relevant lesions are manually marked by ophthalmologists,which takes lots of time and depends on subjective factors excessively.Therefore,it is important to develop fast and accurate algorithm for segmenting retinal image lesions.Based on the above problems,main work of this paper is as follows:(1)An OCT image denoising algorithm is proposed based on self-similarity av-eraging of sampled blocks.The self-similarity prior of image is of great significance for image denoising.The key to the application of image self-similarity lies in the matching of non-local similarity blocks,which is time-consuming.Therefore,we use block downsampling to accelerate the matching process of similar blocks.We exploit weighted averaging of similar blocks to despeckle speckle noise.The experimental results show that the algorithm based on block downsampling has fast operation speed and a better effect of speckle removal.(2)High-order singular value decomposition based on soft threshold and mul-tichannel low-rank tensor denoising are proposed in this chapter.Image block vectorization in low-rank image denoising based on singular value decomposition destroys the topological structure of image blocks,which is not conducive to maintaining image details.We extend two-dimensional low-rank approximation technique to three-dimension.Therefore,a soft threshold despeckling algorithm is proposed by combining the high order singular value decomposition with a specific initialization threshold.In addition,we apply the Eckar-young-mirsky theorem to determine the rank of tensor expansion with noise adaptivity,and the estimated rank is used to shrink the kernel tensor coefficients.Experimen-tal results show that the proposed algorithm has better performance in speckle removal and protection of OCT image details.(3)Windowed active contour model is proposed based on imaging prior and manual interaction.Rapid and accurate retinal OCT image segmentation is very import,ant for clinical diagnosis.OCT image speckle noise is serious and the gray value of lesion is similar to the background.In order to segment retinal OCT im-age lesions rapidly and accurately,we exploitimaging prior of retinal OCT image for interactive windowed segmentation.The interactive windowed segmentation based on imaging prior can enhance the accuracy based on gray value.Exper-imental results show that the proposed interactive windowed segmentation can realize effective segmentation of retinal OCT lesions.In similar low-rank image denoising and windowed interactive segmentation models,the image self-similarity prior and the retinal lesion imaging prior are fully utilized to achieve better image denoising and segmentation results respectively.The research in this paper further strengthens the collaborative innovation of computational mathematics and information science,and deepens and enriches the research on retinal OCT image denoising and segmentation.