Gradient Domain Guided Image Filtering(梯度域导向滤波)

简介: 作者提出了一种新的梯度域引导图像滤波器,通过将明确的一阶边缘感知约束结合到现有的引导图像滤波器中。       matlab代码实现转载至:https://blog.

作者提出了一种新的梯度域引导图像滤波器,通过将明确的一阶边缘感知约束结合到现有的引导图像滤波器中。

 

 

 

 

 

 

 

matlab代码实现

转载至:https://blog.csdn.net/majinlei121/article/details/50717777

%主程序
function
q = gradient_guidedfilter(I, p, eps) % GUIDEDFILTER O(1) time implementation of guided filter. % % - guidance image: I (should be a gray-scale/single channel image) % - filtering input image: p (should be a gray-scale/single channel image) % - regularization parameter: eps r=16; [hei, wid] = size(I); N = boxfilter(ones(hei, wid), r); % the size of each local patch; N=(2r+1)^2 except for boundary pixels. mean_I = boxfilter(I, r) ./ N; mean_p = boxfilter(p, r) ./ N; mean_Ip = boxfilter(I.*p, r) ./ N; cov_Ip = mean_Ip - mean_I .* mean_p; % this is the covariance of (I, p) in each local patch. mean_II = boxfilter(I.*I, r) ./ N; var_I = mean_II - mean_I .* mean_I; %weight epsilon=(0.001*(max(p(:))-min(p(:))))^2; r1=1; N1 = boxfilter(ones(hei, wid), r1); % the size of each local patch; N=(2r+1)^2 except for boundary pixels. mean_I1 = boxfilter(I, r1) ./ N1; mean_II1 = boxfilter(I.*I, r1) ./ N1; var_I1 = mean_II1 - mean_I1 .* mean_I1; chi_I=sqrt(abs(var_I1.*var_I)); weight=(chi_I+epsilon)/(mean(chi_I(:))+epsilon); gamma = (4/(mean(chi_I(:))-min(chi_I(:))))*(chi_I-mean(chi_I(:))); gamma = 1 - 1./(1 + exp(gamma)); %result a = (cov_Ip + (eps./weight).*gamma) ./ (var_I + (eps./weight)); b = mean_p - a .* mean_I; mean_a = boxfilter(a, r) ./ N; mean_b = boxfilter(b, r) ./ N; q = mean_a .* I + mean_b; end %子程序boxfilter() [cpp] view plain copy function imDst = boxfilter(imSrc, r) % BOXFILTER O(1) time box filtering using cumulative sum % % - Definition imDst(x, y)=sum(sum(imSrc(x-r:x+r,y-r:y+r))); % - Running time independent of r; % - Equivalent to the function: colfilt(imSrc, [2*r+1, 2*r+1], 'sliding', @sum); % - But much faster. [hei, wid] = size(imSrc); imDst = zeros(size(imSrc)); %cumulative sum over Y axis imCum = cumsum(imSrc, 1); %difference over Y axis imDst(1:r+1, :) = imCum(1+r:2*r+1, :); imDst(r+2:hei-r, :) = imCum(2*r+2:hei, :) - imCum(1:hei-2*r-1, :); imDst(hei-r+1:hei, :) = repmat(imCum(hei, :), [r, 1]) - imCum(hei-2*r:hei-r-1, :); %cumulative sum over X axis imCum = cumsum(imDst, 2); %difference over X axis imDst(:, 1:r+1) = imCum(:, 1+r:2*r+1); imDst(:, r+2:wid-r) = imCum(:, 2*r+2:wid) - imCum(:, 1:wid-2*r-1); imDst(:, wid-r+1:wid) = repmat(imCum(:, wid), [1, r]) - imCum(:, wid-2*r:wid-r-1); end
 
%运行程序

 

clear

I = double(imread('D:\数字图像处理\研究方向\Filter Smooth\images\tulips.png')) / 255;
% if size(I,3)==3
% I=rgb2gray(I);
% end

p = I;
r=16;
eps = 0.8^2; % try eps=0.1^2, 0.2^2, 0.4^2

q_guide(:,:,1)=guidedfilter(I(:,:,1), p(:,:,1), r, eps);
q_guide(:,:,2)=guidedfilter(I(:,:,2), p(:,:,2), r, eps);
q_guide(:,:,3)=guidedfilter(I(:,:,3), p(:,:,3), r, eps);

q(:,:,1) = gradient_guidedfilter(I(:,:,1), p(:,:,1), eps);
q(:,:,2) = gradient_guidedfilter(I(:,:,2), p(:,:,2), eps);
q(:,:,3) = gradient_guidedfilter(I(:,:,3), p(:,:,3), eps);

figure;imshow([I,q_guide,q]);title('原图,引导滤波,改进引导滤波 eps=0.8^2');

目录
相关文章
【PyTorch实战演练】Fast R-CNN中的RoI(Region of Interest)池化详解
【PyTorch实战演练】Fast R-CNN中的RoI(Region of Interest)池化详解
289 1
YOLOv5的Tricks | 【Trick6】学习率调整策略(One Cycle Policy、余弦退火等)
YOLOv5的Tricks | 【Trick6】学习率调整策略(One Cycle Policy、余弦退火等)
3159 0
YOLOv5的Tricks | 【Trick6】学习率调整策略(One Cycle Policy、余弦退火等)
【DRConv】动态区域感知卷积结构可提升卷积的表示能力 | 复现《Dynamic Region-Aware Convolution》
【DRConv】动态区域感知卷积结构可提升卷积的表示能力 | 复现《Dynamic Region-Aware Convolution》
318 1
【DRConv】动态区域感知卷积结构可提升卷积的表示能力 | 复现《Dynamic Region-Aware Convolution》
使用OpenCV实现Halcon算法(2)形状匹配开源项目,shape_based_matching
使用OpenCV实现Halcon算法(2)形状匹配开源项目,shape_based_matching
4309 0
使用OpenCV实现Halcon算法(2)形状匹配开源项目,shape_based_matching
RT-DETR改进策略【卷积层】| ICCV-2023 引入Dynamic Snake Convolution动态蛇形卷积,改进ResNetLayer
RT-DETR改进策略【卷积层】| ICCV-2023 引入Dynamic Snake Convolution动态蛇形卷积,改进ResNetLayer
82 15
RT-DETR改进策略【卷积层】| ICCV-2023 引入Dynamic Snake Convolution动态蛇形卷积,改进ResNetLayer
RT-DETR改进策略【Conv和Transformer】| CVPR-2024 Single-Head Self-Attention 单头自注意力
RT-DETR改进策略【Conv和Transformer】| CVPR-2024 Single-Head Self-Attention 单头自注意力
52 3
RT-DETR改进策略【Conv和Transformer】| CVPR-2024 Single-Head Self-Attention 单头自注意力
RT-DETR改进策略【Conv和Transformer】| ECCV-2024 Histogram Transformer 直方图自注意力 适用于噪声大,图像质量低的检测任务
RT-DETR改进策略【Conv和Transformer】| ECCV-2024 Histogram Transformer 直方图自注意力 适用于噪声大,图像质量低的检测任务
49 0
【YOLOv11改进 - 注意力机制】EMA(Efficient Multi-Scale Attention):基于跨空间学习的高效多尺度注意力
【YOLOv11改进 - 注意力机制】EMA(Efficient Multi-Scale Attention):基于跨空间学习的高效多尺度注意力.EMA(Efficient Multi-Scale Attention)模块是一种高效多尺度注意力机制,旨在提高计算机视觉任务中的特征表示效果。该模块通过结合通道和空间信息、采用多尺度并行子网络结构以及优化坐标注意力机制,实现了更高效和有效的特征表示。EMA模块在图像分类和目标检测任务中表现出色,使用CIFAR-100、ImageNet-1k、MS COCO和VisDrone2019等数据集进行了广泛测试。
【YOLOv11改进 - 注意力机制】EMA(Efficient Multi-Scale Attention):基于跨空间学习的高效多尺度注意力
【YOLOv8改进】MSFN(Multi-Scale Feed-Forward Network):多尺度前馈网络
**HCANet: 高光谱图像去噪新方法**\n混合卷积与注意力网络(Hybrid Convolutional and Attention Network)是针对HSI去噪的创新模型,结合CNN和Transformer,强化全局与局部特征。它使用卷积注意力融合模块捕获长距离依赖和局部光谱相关性,多尺度前馈网络提升多尺度信息聚合。代码可在[GitHub](https://github.com/summitgao/HCANet)获取。
【YOLOv8改进】 ODConv(Omni-Dimensional Dynamic Convolution):全维度动态卷积
ODConv是一种增强型动态卷积方法,通过多维注意力机制在卷积的四个维度上学习互补注意力,提升轻量级CNN准确性和效率。与现有动态卷积不同,ODConv覆盖了空间、输入/输出通道和核数维度。在ImageNet和MS-COCO上,对MobileNetV2|ResNet等模型有显著性能提升,减少参数的同时超越传统方法。代码和论文链接可用。在YOLO系列中,ODConv改进了特征学习,优化了目标检测性能。