| 汤漫,杨斌.基于快速残差插值和卷积神经网络的去马赛克算法[J].南华大学学报(自然科学版),2019,33(6):68~76.[TANG Man,YANG Bin.Efficient Demosaicking Algorithm Based on Residual Interpolation and Convolution Neural Network[J].Journal of University of South China(Science and Technology),2019,33(6):68~76.] |
| 基于快速残差插值和卷积神经网络的去马赛克算法 |
| Efficient Demosaicking Algorithm Based on Residual Interpolation and Convolution Neural Network |
| 投稿时间：2019-09-30 |
| DOI： |
| 中文关键词: 去马赛克 卷积神经网络 残差插值 全彩色图像 |
| 英文关键词:demosaicking convolutional neural network residual interpolation full-color image |
| 基金项目:国家自然科学基金项目(61871210);湖南省自然科学基金项目(2016JJ3106);湖南省教育厅优秀青年项目(16B225;YB2013B039);南华大学青年英才支持计划、南华大学船山人才支持计划、南华大学重点学科项目(NHXK04) |
|
| 摘要点击次数: 585 |
| 全文下载次数: 370 |
| 中文摘要: |
| 针对传统插值法存在的图像细节不能很好恢复的不足,利用卷积神经网络作为残差插值法的后处理操作,提出了一种基于残差插值和卷积神经网络的去马赛克算法。方法分为初始去马赛克和细节恢复后处理两部分。先用改进的基于梯度的快速残差插值法实现初步去马赛克插值,并针对恢复图像中包含了彩色伪影,细节丢失等问题,再使用深度残差网络学习恢复图像与理想全彩色图像之间的映射,实现后处理。在Kodak数据集和IMAX数据集上的实验结果表明,该方法结果在主观视觉特性和客观评价指标两方面相较于传统方法都有明显改进。 |
| 英文摘要: |
| Aiming at the insufficiency of the image details in the traditional interpolation method,this paper proposes a demosaicking algorithm based on residual interpolation and convolutional neural network by using the convolutional neural network as the post-processing operation of residual interpolation.The method is divided into two parts:initial demosaicking and detail recovery post-processing.Firstly,the initial demosaick interpolation is implemented by the improved gradient-based fast residual interpolation method,and the color image is included in the restored image,and the details are lost.Then,the deep residual network is used to learn the relationship between the restored image and the ideal full-color image.The mapping is implemented post-processing.The experimental results on the Kodak dataset and the IMAX dataset show that the results of the proposed method are significantly improved compared with the traditional methods in subjective visual characteristics and objective evaluation. |
| 查看全文 查看/发表评论 下载PDF阅读器 |
| 关闭 |
|
|
|