基于生成对抗网络的色盲测试图像自动生成方法研究

Research on Automatic Generation Method of Color Blindness Test Image Based on Generative Adversarial Networks

色觉作为人眼重要的视功能之一,已成为眼部体检中的一项常规检查项目,临床中最常用最常见的检查方法就是通过色盲测试图测试。然而,设计色盲测试图像非常复杂,浪费人力、物力,从而降低了色觉检查的准确率和效率。因此,基于生成对抗网络(Generative Adversarial Networks,GAN)设计一种自动生成色盲测试图像的方案。首先在网络上收集数字0-9的色盲测试图像,其次利用卷积神经网络设计生成器和判别器,最后通过生成器和判别器形成互相博弈的态势,以优化生成器生成足够逼真的色盲测试图像。实验结果表明,该方法较好地解决了色盲测试图像设计的难题,从而帮助医学检查者实施更加准确的色觉异常判定。

As one of the important visual functions of the human eye,color vision has become a routine examination item in the eye physical examination,and the most commonly used and common exma ination method in clinical is to pass the color blindness tes cthart.However,the design of color blindness test images is very complicated,which wastes manpower and material resources,thus reducing the accuracy and efficiency of color vision inspection.Therefore,based on Generative Adversarial Network(GAN),this study designed a scheme to automatically generate test images for color blindness.Firstly,0-9 color blindness test images are collected onth e network.Secondly,a generator and a discriminator are designed by using convolutional neural network.Finally,a mutual game between the generator and the discriminator is formed to optimize the geneartor to generate a sufficiently realistic color blindness testim age.The experimental results show that this method can solve the problem of color blindness test image design,and help medical examiners to determine color vision abnormality more accurately.