查询结果:   李晓戈,薛倩茹.基于深度卷积神经网络的图像去雾算法[J].计算机应用与软件,2019,36(8):189 - 195.
中文标题
基于深度卷积神经网络的图像去雾算法
发表栏目
图像处理与应用
摘要点击数
368
英文标题
IMAGE DEHAZING ALGORITHM BASED ON DEEP CONVOLUTIONAL NEURAL NETWORK
作 者
李晓戈 薛倩茹 Li Xiaoge Xue Qianru
作者单位
西安邮电大学计算机学院 陕西 西安 710121     
英文单位
School of Computer Science and Technology, Xi’an University of Posts and Telecommunications, Xi’an 710121,Shaanxi, China     
关键词
图像去雾 卷积神经网络 信息叠加 高频 低频
Keywords
Image dehazing Convolutional neural network Information superposition High frequency Low frequency
基金项目
陕西省科技重点研发计划项目与咸阳市科技研究计划项目联合支持项目(2018ZDXM-GY-043)
作者资料
李晓戈,教授,主研领域:自然语言处理,深度学习。薛倩茹,硕士生。 。
文章摘要
随着人们对图像的质量要求越来越高,相比于传统的去雾算法,人们发现用卷积神经网络(Convolutional Neural Networks,CNN)对图像进行去雾处理可以达到更好的效果,可以更好地还原图像的轮廓和细节。通过研究CNN去雾的原理,提出一种通过深度卷积神经网络对图像进行去雾处理的模型。用该算法得到图像的高频信息与去雾前的低频信息相叠加,以得到清晰的图像。将该算法和基于模型和基于神经网络的去雾最新算法进行对比,实验结果表明,该算法在峰值信噪比(Peak Signal to Noise Ratio,PSNR)和时间上都优于其他几种算法,并且在细节处理和图像纹理恢复上效果也更好。
Abstract
With the increasing demand for image quality, when compared with traditional dehazing algorithms, we find that the use of CNN(Convolutional Neural Networks) to dehaze the image can achieve better results, and restore the contours and details of the image. By studying the principle of CNN dehazing, a model of image dehazing by deep convolution neural network was proposed. The high-frequency information of the image obtained by the new algorithm was superimposed with the low-frequency information before dehazing to obtain a clear image. And the new algorithm was compared with model-based and convolutional neural network-based algorithms. The experimental results show that the new algorithm is superior to other several algorithms in PSNR(Peak Signal to Noise Ratio) and time, and has better effect in detail processing and image texture restoration.
下载PDF全文