Retina Recognition Based On PCA-SIFT Operator

Objective: Retina is a kind of fixed biological characteristics with strong stability and accuracy.Retina has a high degree of security can be widely used in identity recognition.This paper aims to study and design an adaptive threshold segmentation algorithm for optic disc,through which the accurate coordinate position and radius of optic disc are located in retinal images,so as to facilitate auxiliary retinal recognition and subsequent application of fundus images to optic disc.In this paper,two retinal recognition algorithms are designed for rotation,brightness and other factors.By adding rotation,brightness and noise factors,the performance of the retinal algorithm is tested and whether the algorithm has the invariance of the corresponding factors is verified.Methods: In the study of optic disc segmentation,this paper mainly proposed an adaptive threshold optic disc segmentation algorithm aiming at the high brightness effect of retinal images caused by lesions: in the pre-experiment,the blood vessels were first eliminated to reduce the influence of blood vessels on the location of optic disc.On this basis,the double Otsu threshold method is used to segment the optic disc.In this paper,the selection of the target region is optimized,and a threshold detection of the optic disc area is proposed.Finally,Hof transform is used to locate the center coordinates of the optic disc and the radius of the optic disc.In the research of retinal recognition,this paper uses the PCASIFT(Principal component analysis-Scale invariant feature transform)operator to describe the features of retinal extraction.In order to enable the PCA-SIFT operator to better extract the feature points on blood vessels,this paper proposes the optimization of blood vessel enhancement to enhance the contrast of blood vessels.While retaining the features of SIFT(Scale invariant feature transform)algorithm,PCA(Principal component analysis)is used to reduce the dimension of feature points.In this paper,40 fundus images in DRIVE database and 1200 fundus images in MESSIDOR database were used as experimental data.Results: In the experiment of DRIVE database,the accuracy rate of optic disc segmentation was 97.5%,and the accuracy rate of retina matching based on PCA was97.5%.The accuracy of retina matching based on PCA-SIFT operator is 100%.In the experiment of MESSIDOR database,the accuracy rate of optic disc segmentation was98.75%,the seriousness rate of retina matching based on PCA was 97.17%,and the seriousness rate of retina matching based on PCA-SIFT operator was 97.83%.At the same time,the influence of rotation,brightness and noise factors on the two algorithms is designed to compare the experiment.Conclusion: In the optic disc segmentation part,the algorithm of this paper has a more significant segmentation effect on the retina image containing diseased retina,and for the optic disc fuzzy image has a good adaptive ability,in the retina recognition part,the conscientious rate is higher than other algorithms,which in the DRIVE database experiment,the conscientious rate of this algorithm is significantly higher than other algorithms.In the comparison test,the two algorithms in this paper have a certain degree of rotation invariance,brightness invariance,and a certain degree of noise resistance.