Fundus image completion and classification method and system

Abstract

The invention discloses a fundus image completion and classification method and system, and the method comprises the steps: carrying out the noise detection of a fundus image, and marking a noise region; extracting blood vessel direction features in the fundus image; complementing the noise region by taking the extracted blood vessel direction features as constraints; performing image classification on the completed fundus image through a pre-trained fundus image classification model, and outputting recognition and classification results of the fundus image. According to the invention, the blood vessel direction features are extracted, and the noise region on the fundus image is complemented by taking the blood vessel direction features as constraints, so that the damaged fundus image is well restored; an eye fundus image classification model is formed through improved convolutional residual network training to recognize and classify eye fundus images, blood vessel features on the eye fundus images can be rapidly and accurately extracted, and the eye fundus image classification efficiency and accuracy are improved.

Description

technical field [0001] The invention relates to the field of image processing, in particular to a fundus image completion and classification method and system. Background technique [0002] Fundus images contain rich vascular features. However, due to the influence of the external environment of image acquisition or the presence of dirt in the eye, there will be noise in the collected fundus image. These noise areas will affect the extraction of blood vessel features, resulting in inaccurate fundus image classification, and the fundus classification result lacks reference significance. [0003] In addition, there are two main existing methods for fundus image classification: [0004] One is to manually classify fundus images by doctors based on their experience. The classification results of this fundus image classification method are usually more accurate, but the classification process is heavily dependent on the personal experience of doctors and cannot be popularized on...

AI Technical Summary

Problems solved by technology

This method automates the recognition and classification of fundus images. However, since the blood vessels in fundus images are usually very small, traditional machine learning recognition methods cannot accurately identify the characteristics of blood vessels, and it is difficult to identify the differences between the characteristics of blood vessels in different fundus images. The results of fundus image classification are of little reference value

Images