我正在尝试基于此 MATLAB 代码制作拉普拉斯金字塔:
% I is the image array, where its size is r x c x 3(<-RGB channels).
function pyr = laplacian_pyramid(I,nlev)
r = size(I,1);
c = size(I,2);
% recursively build pyramid
pyr = cell(nlev,1);
filter = [.0625, .25, .375, .25, .0625];
J = I;
for l = 1:nlev - 1
% apply low pass filter, and downsample
I = downsample(J,filter);
odd = 2*size(I) - size(J); % for each dimension, check if the upsampled version has to be odd
% in each level, store difference between image and upsampled low pass version
pyr{l} = J - upsample(I,odd,filter);
J = I; % continue with low pass image
end
pyr{nlev} = J; % the coarest level contains the residual low pass image
其中,downsample()
看起来像这样:
function R = downsample(I, filter)
border_mode = 'symmetric';
% low pass, convolve with separable filter
R = imfilter(I,filter,border_mode); %horizontal
R = imfilter(R,filter',border_mode); %vertical
% decimate
r = size(I,1);
c = size(I,2);
R = R(1:2:r, 1:2:c, :);
这里是upsample()
:
function R = upsample(I,odd,filter)
% increase resolution
I = padarray(I,[1 1 0],'replicate'); % pad the image with a 1-pixel border
r = 2*size(I,1);
c = 2*size(I,2);
k = size(I,3);
R = zeros(r,c,k);
R(1:2:r, 1:2:c, :) = 4*I; % increase size 2 times; the padding is now 2 pixels wide
% interpolate, convolve with separable filter
R = imfilter(R,filter); %horizontal
R = imfilter(R,filter'); %vertical
% remove the border
R = R(3:r - 2 - odd(1), 3:c - 2 - odd(2), :);
此 MATLAB 代码运行正常。只关注 downsample()
函数,因为我的 OpenCV 版本的那个函数就是我的问题发生的原因。
现在这是我对这个 MATLAB 代码的 OpenCV 版本的尝试:
private List<Mat> laplacianPyramid(Mat mat,int depth)
{
//mat.type() is CV_8UC3 (16).
List<Mat> pyramid = new ArrayList<Mat>();
//I make a clone so I don't ruin the original matrix.
Mat clone = mat.clone();
Mat J = clone;
for(int i=0;i<=depth-2;i++)
{
clone = image_reduce(J);
Mat temp = new Mat();
Point odd = new Point(clone.size().height*2 - J.height(), clone.size().width*2 - J.width());
Core.subtract(J, image_expand(clone, odd), temp);
pyramid.add(temp);
J = clone;
}
pyramid.add(J);
return pyramid;
}
这是我的 upsample()
的 OpenCV 版本:
private Mat image_expand(Mat image, Point odd){
//I make a clone so I don't ruin the original image.
Mat imageClone = image.clone();
copyMakeBorder(imageClone, imageClone, 1, 1, 1, 1, BORDER_REPLICATE);
Mat kernelX = getGaussianKernel();
Mat kernelY = new Mat();
Core.transpose(kernelX, kernelY);
Mat UIntVer = new Mat(imageClone.size(), CV_8UC3);
imageClone.convertTo(UIntVer, CV_8UC3);
Imgproc.resize(UIntVer, UIntVer, new Size(imageClone.width()*2, imageClone.height()*2), 0, 0, Imgproc.INTER_NEAREST);
//Now implement the zero padding between each columns and rows, just like the MATLAB version.
Mat mask = new Mat(2,2, CV_8UC1);
int[][] array = new int[2][2];
array[0][0] = 255;
array[1][0] = 0;
array[0][1] = 0;
array[1][1] = 0;
for (int i=0; i<2; i++) {
for (int j = 0; j < 2; j++) {
mask.put(i, j, array[i][j]);
}
}
//mask becomes twice the size of image.
Mat biggerMask = new Mat();
Core.repeat(mask, imageClone.height(), imageClone.width(), biggerMask);
List<Mat> rgbUIntVer = new ArrayList<Mat>();
Core.split(UIntVer,rgbUIntVer);
Core.bitwise_and(rgbUIntVer.get(0), biggerMask, rgbUIntVer.get(0));
Core.bitwise_and(rgbUIntVer.get(1), biggerMask, rgbUIntVer.get(1));
Core.bitwise_and(rgbUIntVer.get(2), biggerMask, rgbUIntVer.get(2));
Core.merge(rgbUIntVer, UIntVer);
int r = imageClone.height()*2;
int c = imageClone.width()*2;
Mat result = new Mat(r, c, CV_32FC3);
UIntVer.convertTo(UIntVer, CV_32FC3);
Scalar four = new Scalar(4);
Core.multiply(UIntVer, four, UIntVer);
Imgproc.sepFilter2D(UIntVer,result,-1,kernelX,kernelY,new Point(-1,-1) ,0,BORDER_DEFAULT);
Rect roi = new Rect(2, 2, c-4-(int)odd.y, r-4-(int)odd.x);
result = new Mat(result, roi);
return result;
}
这个 OpenCV 版本代码的问题是结果只是纯蓝色。 Laplacian Pyramids 应该保存边缘检测结果,但示例结果是这样的。上图是原始输入,下图是生成的金字塔的底层。
我检查了输入图像是否被正确读取。我的代码有问题,但我找不到哪里。任何帮助将不胜感激!
我怀疑某些处理函数只在 R channel 上工作,而不在 GB channel 上工作。
我发现问题出在 image_expand()
上。它只是简单地输出一个红色版本的图像,这解释了为什么结果图像是蓝色的,因为结果图像是原始图像和 image_expand()
的输出之间的差异。所以问题出在 image_expand()
中。
以防万一,这是我的 getGaussianKernel()
代码:
private Mat getGaussianKernel(){
float[] kernel = new float[5];
kernel[0]= 0.0625f;
kernel[1]=0.25f;
kernel[2]=0.375f;
kernel[3]=0.25f;
kernel[4]=0.0625f;
Mat mat = new Mat(5,1,CV_32FC1);
mat.put(0,0,kernel);
return mat;
}
最佳答案
我发现了问题。正如@Ander Biguri 先生和@Cris Luengo 先生(对不起,如果我假设你的性别)告诉我,问题出在
UIntVer.convertTo(UIntVer, CV_32FC3);
Scalar four = new Scalar(4);
Core.multiply(UIntVer, four, UIntVer);
在 image_expand()
中。显然,要乘以标量值,您必须对图像矩阵执行 Core.split()
,并对每个 channel 应用乘法,然后执行 Core.merge()
它。这是因为 Core.multiply()
一次只处理 1 个 channel 。我希望这可以帮助其他人发现 Core.multiply()
的问题。
关于java - 拉普拉斯金字塔产生奇怪的结果?,我们在Stack Overflow上找到一个类似的问题: https://stackoverflow.com/questions/53229340/