基于深度残差学习的乘性噪声去噪方法

Multiplicative Denoising Method Based on Deep Residual Learning

图像去噪是数字图像处理中最基本的研究内容,也是一项十分关键的技术,一直以来是图像处理领域的难点。图像去噪的好坏直接影响后续图像边缘检测、特征提取、图像分割和模式识别等图像处理。为有效去除乘性噪声的影响,提出一种深度残差学习的乘性噪声去噪方法。该方法通过引入残差优化,解决了卷积神经网络在层数较多时,随着层数加深,梯度在传播过程中逐渐消失的问题。与4种经典去噪算法进行比较,结果表明,该方法在有效去除乘性噪声的同时,可以更好地保留图像的边缘和纹理区域的细节信息,为后续的图像分割、配准和目标识别等奠定基础。

Image denoising is the most basic problem and a key technology in digital image processing, which has always been difficult in the field of image processing. The quality of image denoising directly affects the follow-up image processing, such as image edge detection, feature extraction, image segmentation, and pattern recognition. In order to effectively remove the influence of multiplicative noise, we propose a denoising method based on deep residual learning, which solves the problem that the gradient gradually disappears when the number of convolutional neural network＇s layers increases by residual optimization. By comparing with four classical denoising algorithms, we make the conclusions that the proposed method can not only effectively remove the multiplicative noise, but also preserve the edge of the image and the detail information of the texture area, which will lay the foundation for image segmentation, registration, object recognition, and so on.