基于深度多尺度卷积网络Autoencoder的数字图像去噪算法
首发时间:2016-11-14
摘要:图像去噪是从带噪声的图像中恢复出原始图像的问题。本文提出了一种基于深度多尺度卷积网络的Autoencoder(deep multi scale convolutional neural networks,DM-CAE),能够达到先进的良好去噪效果。更重要的是,传统的去噪算法通常只对一种类型的噪声有效,该方法对于多种噪声均具有有效性和健壮性。我们在MNIST数据集上进行实验,实验结果证明,对于多种噪声,该方法均具有健壮性和有效性。将此算法作为预处理算法,能够显著提高带噪声图像的分类准确率。
关键词: 计算机应用 图像去噪 卷积神经网络 Autoencoder
For information in English, please click here
DM-CAE for Image Denoising
Abstract:Image denoising is a problem recovering original noisy-free image from a noisy image.In this paper, we present a novel method that combine autoencoder with deep multi scale convolutional neural networks(DM-CAE),can achieve comparable performance to the current state-of-the-art image denoising methods.More importantly, traditional methods are just useful to one type of noise, but our method is robust to various types of noise. We demonstrate our method as a denoise algorithm and a preprocessing architecture of the classification on the MNIST dataset. Experimental results demonstrate the effectiveness and robustness of the proposed method in the tasks of image denoising and performance of improving the classification accuracy on corrupted digit images.
Keywords: computer application technology image denoise convolutional neural network Autoencoder
基金:
论文图表:
引用
No.4709005117032914****
同行评议
共计0人参与
勘误表
基于深度多尺度卷积网络Autoencoder的数字图像去噪算法
评论
全部评论0/1000