基于卷积自编码的图像降噪研究

Research on image denoising based on convolutional autocoding

当今时代图像信息占的比例越来越重,图像在收集和传送中会产生各种噪声,导致图像的质量变得很差,所以图像的降噪的任务非常关键,常见的有椒盐噪声和高斯噪声,常用中值滤波和均值滤波传统的方法来去除,卷积自编码器进行降噪,它属于神经网络的一种无监督学习算法。通过卷积和池化操作技术实现图像信息的特征提取。在降噪卷积自编码器模型的输入端传入一个带有不同程度的高斯噪声图像,测试它降噪的效果。在输出端输入原始图像,运用梯度下降算法来不断地迭代训练网络模型,使自编码器学到其中的规律,迭代了100次之后训练的模型最佳降噪效果最好。

Nowadays,the proportion of image information is more and more heavy.The image in the collection and transmission can produce all kinds of noise,lead to the quality of the image becomes very poor,so the task of image noise reduction is critical,common salt and pepper noise and gaussian noise,median filter and mean filter is commonly used to remove to the traditional methods,the main research content of this article is to use the convolution encoder for noise reduction,It belongs to an unsupervised learning algorithm of neural network.Feature extraction of image information is realized by convolution and pooling.A gaussian noise image with different degrees is introduced into the input end of the denoising convolutional autoencoder model to test its denoising effect.The original image is input at the output end,and the gradient descent algorithm is used to continuously iterate the training network model so that the autoencoder can learn the rules.After 100 iterations in this paper,the trained model has the best noise reduction effect.