摘要
图像生成是虚拟现实技术(virtual reality,VR)中的重要技术手段,针对传统图片生成方法需要大量的数据集进行训练,且生成的图片轮廓不清晰等问题,采用基于深度卷积神经网络和生成对抗网络来实现图片的生成。为了保证生成图片的真实性和完整性,在图片生成阶段引入变分自编码器,通过编码器获取到输入图片数据的均值和方差,将图片对应的隐藏变量转化为标准的高斯分布,然后通过生成器生成新的图片;在识别阶段,采用深度卷积神经网络训练判别器,将生成的新的图片输入到已经训练好的判别器中,运用梯度下降法计算损失函数,不断优化整体系统模型。通过对MNIST图像数据集的训练,实验表明该方法能生成质量较高的图片,它生成的图像无法用肉眼与真实数据区分开,并且在不同网络条件下都有较高的识别率。该方法提高了MNIST生成模型的技术水平。
Image generation is an important part of virtual reality(VR).In order to solve the problem that traditional image generation method needs a large number of data sets for training and the generated image contour is not clear,the deep convolutional neural network and generation adversation network are used to realize the image generation.To ensure the authenticity and integrity of the generated image,a variational auto-encoder is introduced in the image generation stage.The mean value and variance of the input image data are obtained by the encoder,and the hidden variables corresponding to the image are transformed into the standard Gaussian distribution,then the new image is generated by the generator.In the recognition stage,the deep convolutional neural network is used to train the discriminator,and the generated new images are input into the trained discriminator.The gradient descent method is used to calculate the loss function and continuously optimize the overall system model.Through the training of MNIST image data set,the experiment shows that the proposed method can generate high-quality images which cannot be distinguished from the real data with the naked eye,with high recognition rate under different network conditions.It improves on the state of the art for generative models on MNIST.
作者
尹玉婷
肖秦琨
YIN Yu-ting;XIAO Qin-kun(School of Electronic Information Engineering,Xi’an Technological University,Xi’an 710021,China)
出处
《计算机技术与发展》
2021年第4期86-92,共7页
Computer Technology and Development
基金
国家自然科学基金资助项目(61671362,6207010855)。
关键词
生成对抗网络
深度卷积网络
变分自编码器
图像生成
梯度下降法
generative adversarial network
deep convolutional network
variational auto-encoder
image generation
gradient descent method
