Research On Algorithm Of Diabetic Retinal Image Vessels Segmentation

Many diseases are closely related to changes in retinal vascular morphology,such as diabetes,hypertension,cardiovascular and other common diseases,of which diabetic retinopathy has the largest proportion.Traditional artificial detection of retinal blood vessel images is labor intensive and time consuming.Although doctors can segment retinal blood vessel images with high precision,excessive effort can lead to an increase in the workload of doctors.With the widespread use of digital image processing technology in the field of medical imaging,the automatic segmentation of retinal blood vessels on computer systems has become a trend.At present,researchers have proposed a number of algorithms in retinal blood vessel extraction,and have achieved certain results,but the impact on the structure of the retina and the tissue of the lesion needs to be further improved.This article has conducted in-depth research on retinal vascular image enhancement and segmentation techniques.The main work of the thesis is as follows:In terms of retinal image enhancement,this paper proposes a method combining Gabor transform with high and low hat transform.Gabor has the characteristics of frequency selection and direction selection.It can also suppress the noise by setting the threshold of high frequency subband coefficient,and the high and low cap transforms the optic disc.Both the lesion and the lesion area have been treated to suppress the influence of the highlight area and the dark area.In this paper,the advantages of the two are combined,firstly,the Gabor transform is performed on the retinal blood vessel image,and the low-frequency cap image is processed in the obtained low-frequency image.In the frequency image,the threshold analysis method is adopted,and finally the reconstruction is performed.The contrast of the blood vessel and the background in the enhanced image is obviously improved,and the characteristics of the small blood vessels can be highlighted,and the experiment is compared with the common enhancement method.The enhancement method of this paper can effectively improve the contrast between blood vessels and background.In the aspect of retinal vascular image segmentation,this paper proposes a threshold segmentation method combining multi-scale linear detection and improved two-dimensional maximum entropy threshold.Multi-scale linear detection does not correctly segment the optic disc and the blood vessels in the lesion tissue area.The retinal image enhancement algorithm combined with the Gabor transform and the high and low cap transform proposed in this paper solves this problem well.The multi-scale linear detection algorithm needs to constantly adjust the threshold parameters to obtain the threshold.In this paper,the two-dimensional maximum entropy threshold algorithm is implemented by the flower pollination algorithm.Optimization,automatic selection of optimal threshold,the algorithm is simple and easy to implement,with high precision.Finally,the retinal blood vessel image is segmented according to the obtained optimal threshold.Experiments show that the segmentation method of this paper has the advantage of segmenting more small blood vessels.In this paper,the diabetic retinal blood vessel image in the DRIVE database was selected on the MATLAB platform.The experimental results were compared with the results of expert manual segmentation and several typical segmentation algorithms.It was concluded that the segmentation method proposed in this paper solved the optic disc and lesion area.The problem of being mis-segmented can also segment more small blood vessels.Finally,the effectiveness of the proposed algorithm is further proved by objective evaluation data.