Research On Retinal Image Segmentation Algorithm Based On Texture Features

The blood vessels in the retinal image are a kind of microvessels that can be observed.The changes of their distribution,structure and morphological characteristics can reflect the degree of pathological changes to a certain extent.With the development of medical imaging technology,it is not advisable to only rely on artificial diagnosis of retinal diseases,and computer-aided diagnosis and treatment has become the most common means.It has great significance in the diagnosis of diabetic retinopathy,glaucoma and optic neuritis.Blood vessels segmentation and optic disc location are two main tasks related to retinal image analysis.Because of the great difference in the environment of obtaining the retinal database,the uneven illumination,the imbalance of contrast and the bright spots and bleeding of the lesions will increase the difficulty of retinal image processing.The optic disc is the most important physiological structure in the fundus image.Accurate location of the optic disc is conducive to accurately determine the cause of fundus lesions In the pathological image,the optic disc may be completely covered,or part of it may be covered by bright spots,or there is no optic disc in the image.In normal images,the optic disc is the brightest area,it is easy to be confused with large bright yellow lesions.If the optic disc area can be accurately detected,it can be screened out from the candidate lesions.For the problem of optic disc location,this paper designs a method of optic disc location based on texture feature and shape feature.This method is robust to both normal and pathological fundus images.Compared with other methods,this algorithm has some advantages in the accuracy and calculation efficiency,and can provide the basis for the next work of blood vessel segmentation.For the problem of blood vessel segmentation,this paper designs a method of blood vessel segmentation based on multi texture feature theory.This method is based on texture feature quantization,which is different from other blood vessel segmentation methods.We have found a large number of texture features that can be used in image processing,and collected 41 texture features.Through the quantitative analysis of the blood vessel texture features of the retinal image,we found that several texture features,such as standard deviation features,can easily identify the foreground and background of the blood vessel,so as to obtain a better visual effect than the original image,which can provide more information for blood vessel segmentation and visual disc positioning.At the same time,we design a retinal blood vessels segmentation algorithm which combines the energy texture feature,standard deviation texture feature,morphological operation and the maximum class difference method.The results show that the method of multi feature fusion is more effective than that of single feature segmentation,This method improves the accuracy of blood vessel segmentation,makes the structure of blood vessels more complete,and solves the problem of loss of optic disc area and surrounding blood vessels in the process of segmentation to a certain extent.