Feature extraction method for kernel type cataract image

Abstract

The invention discloses a feature extraction method for a kernel type cataract image. The feature extraction method comprises the following steps: step 1, carrying out fuzzy processing on a clear kernel type cataract image to obtain a corresponding clear-fuzzy image pair data set, training a DebuurGAN model by using the clear-fuzzy image pair data set, and removing blurring of an actual collectedimage by using the model to obtain a deblurred actual collected kernel type cataract image data set; 2, training a CNN model according to the deblurred actually collected nuclear cataract image data set, and obtaining a corresponding image feature extractor according to the trained CNN model; 3, generating a clear image from the kernel type cataract image with the to-be-extracted characteristics by using the trained DebuurGAN model, and calculating the obtained clear image by using an image characteristic extractor to obtain the characteristics of the image. According to the method, the problem of interference of nuclear cataract fuzzy weak samples in the medical field on feature extraction can be relieved, a large number of clear samples do not need to be manually collected to support model learning, and the robustness is better.

Description

technical field [0001] The invention belongs to the technical field of computer vision processing, and in particular relates to a feature extraction method for nuclear cataract images. Background technique [0002] Computer vision technology has been widely used in fields such as medicine, industry, agriculture, transportation, and daily life. It can be well qualified for image feature recognition and subsequent processing, and provides great help for the intelligent and fast processing of various information. . [0003] Among them, in image feature extraction, when collecting images, due to the influence of various complex environments, some images are taken in dark environments, and some images are taken in bright environments. Images from sources under different lighting and shooting situations have different specific performances of features, and there are different interference factors. The heterogeneous problems include light leakage in the dark environment, which le...

AI Technical Summary

Problems solved by technology

[0009] The purpose of the present invention is: in the current application of computer image feature extraction, multi-source heterogeneous interference and small samples of sample data when collecting images in medical and other fields cause the problem of fuzzy and weak samples, which seriously interferes with image feature recognition methods. Moreover, the existing image feature extraction methods need a large number of clear samples to support the learning of the model, which is difficult to apply to the situation of fuzzy and weak samples. In order to solve the above problems, a feature extraction method for nuclear cataract images is proposed.

Images