A Research On Structure Recognition Of Fundus Image

It is well known that eye examination plays an important role in the diagnosis and treatment of most eye diseases and part of body diseases.Fundus images are indispensable for eye examination,they can reflect the real situation of patients objectively and help doctors make diagnosis.At present using computer technology to process the captured fundus images is a necessary step in clinical eye examination.It could help doctors observe the fundus images more quickly and accurately,so as to improve the accuracy and efficiency of diagnose.This step is the basis to assist doctors to diagnose and treat eye diseases,it has important significance.In this paper,we study the structure recognition of fundus,which is one of the common fundus images process jobs.This paper presents an optic disk and macula recognition method which uses image registering techniques and color label transfer idea.The main idea of the method is registering the processed fundus image and the reference fundus images with known structures,then transferring the color labels from the reference fundus images with known structures to the processed fundus image,and after that,optimizing the above results to obtain the accurate color labels of the processed fundus image.The obtained color labels are indeed the results of structure recognition of the processed fundus image.In this research,we use four databases(DRIVE,DIARETDB0,DIARETDB1 and STARE)which are commonly used by scholars in the studies of fundus images to complete the experiment of the method proposed in this paper.We find that this method could be used to recognize optic disk and macula in fundus image and obtain good results,especially for the macula recognition problem in fundus image.In most of existing macula recognition methods,researchers use the result of optic disk recognition to recognize macula.Here in this paper,we propose a new method,in which we recognize macula directly from fundus images.Without the limitation of theoptic disk recognition result,this method ensures the high accuracy and stability of macula recognition in fundus image.In this paper,we designed four indexes to quantitatively evaluate and analyze the experimental results.From the analysis results,we find that the method proposed in this paper has high recognition rate and accuracy on structure recognition of fundus.The accuracy of this method is obviously higher than other methods.Through the analysis of the structure recognition of fundus images without lesions and fundus images with lesions,we find that this method has good stability.But in the case of fundus images with severe lesions,the results of recognize optic disk are not enough good.The time required for the structure recognition of fundus images is longer than that of other methods,which shows that this method needs to be improved further in computational efficiency.