Vessel Segmentation Based On Wavelet Transform And Steerable Gaussian Filter In Fundus Image

Retinal vessels are the only deep microvessels that can be directly observed without trauma in the human body.They are the observation window for many fundus diseases such as glaucoma and systemic diseases such as diabetic retinopathy and hypertension.Fundus color photography is a common technique to obtain fundus blood vessels in ophthalmology.The segmentation of blood vessels in fimdus color photography is very important in the diagnosis and treatment of ophthalmology and other diseases.The automatic detection and accurate analysis of retinal vessels can help doctors improve the screening efficiency of diabetic retinopathy and other diseases.But the automatic segmentation of blood vessels in retinal images of fundus is very challenging.(2)there are many lesions in the pathological retina;(3)the color contrast of retinal fundus was not uniform,and the contrast between blood vessels and background was low.Based on the analysis of the status of retinal vessel segmentation in the world,this paper proposes an extraction algorithm for retinal vessels based on the combination of wavelet transform and steerable gaussian filter.Firstly,the green channel of the color image on the fundus was vascularized by using isotropic undedmated wavelet transform,and then the enhaneed image was binarized.Then the binarization image was filtered and mask removed by morphological etching,and the candidate vessel I1 was obtained.Then the vessels in the green channel of the color image under the eye base were enhanced in multiple directions based on the Steerable gaussian filter,and the eandidate vessels 12 were obtained through binarization processing.Finally,I1 and 12 were logic and operated to obtain the final results of retinal vessel segmentation.The algorithm was validated on DRIVE and STARE base image database.The accuracy rate on DRIVE database is 95.49%,sensitivity is 77.87%,specificity is 96.97%,accuracy rate on STARE database is 95.01%,sensitivity is 75.67%and specificity is 96.89%.The results show that the combination of these two methods can effectively improve the accuracy of vessel segmentation.The results of comparison with other vessel segmentation algorithms show that the performance of this algorithm is very competitive.