Author(s)

Hang Zhou, Wei Dong, Zhenya Zhang, and Yinghui Jiang

Corresponding Author

Hang Zhou

ABSTRACT

In order to reduce the workload of medical workers, improve work efficiency, reduce errors, this paper studied a can learn human retinal fundus image difference system, for the purpose of this vision system in the application of fundus figure, can automatically determine the difference between the eye diagram of the different people and make a basic analysis, in order to do the preliminary screening of disease, whether the person will have the risk of diabetes, high blood pressure, etc. This paper makes the following contents based on image processing technology. First browsed the traditional visual algorithm is relatively simple in the SIFT (Scale-invariant feature transform) algorithm, which is a computer vision algorithm, used to detect and describe the local characteristics of image, it seeks extreme value point in the space Scale, and extract its location, Scale and rotation invariant, eventually fundus figure by the standards of the already sick with the generation of mapping similarity comparison, to determine the generation of fundus photographs himself at risk. Secondly, after recognizing the shortcomings of the traditional algorithm, such as the weak anti-interference ability for outliers, sometimes fewer feature points, and inability to accurately extract feature points for targets with smooth edges, etc. This paper finally uses CNN deep learning algorithm and successfully realizes it in retinal fundus image processing. The results show that the application of the retinal fundus image in the CNN algorithm can effectively eliminate the interference of abnormal points.Finally, the fundus images of different patients were classified to match the corresponding diseases.

KEYWORDS

retinal fundus image processing, TensorFlow, CNN, disease