opencv split vs mixChannels - opencv

To separate hue channel from HSV image, here is the code using the mixChannels function:
/// Transform it to HSV
cvtColor( src, hsv, CV_BGR2HSV );
/// Use only the Hue value
hue.create( hsv.size(), hsv.depth() );
int ch[] = { 0, 0 };
mixChannels( &hsv, 1, &hue, 1, ch, 1 );
But I know split function can also do this:
vector<Mat> chs;
split(hsv, chs);
Mat hue = chs[0];
Is that OK?
If these are the same, I think split method is more clean. Am I right?

You are pretty much right, split() is used to split all the channels of an multi-channel matrix into single channel matrices. On the other hand if you are interested in only one channel you can use mixChannels(). So you don't have to allocate memory for other channels as we do with split().

Keep things simple and use extractChannel, which wraps mixChannels for you.
cv::Mat hue;
int cn = 0; // hue
cv::extractChannel(hsv, hue, cn);

Related

opencv c++ inverse fourier transformation does not give same image

I have a bgr image and convert to lab channels.
I tried to check if the idft image of the result of dft of L channel image is the same.
// MARK: Split LAB Channel each
cv::Mat lab_resized_host_image;
cv::cvtColor(resized_host_image, lab_resized_host_image, cv::COLOR_BGR2Lab);
imshow("lab_resized_host_image", lab_resized_host_image);
cv::Mat channel_L_host_image, channel_A_host_image, channel_B_host_image;
std::vector<cv::Mat> channel_LAB_host_image(3);
cv::split(lab_resized_host_image, channel_LAB_host_image);
// MARK: DFT the channel_L host image.
channel_L_host_image = channel_LAB_host_image[0];
imshow("channel_L_host_image", channel_L_host_image);
cv::Mat padded_L;
int rows_L = getOptimalDFTSize(channel_L_host_image.rows);
int cols_L = getOptimalDFTSize(channel_L_host_image.cols);
copyMakeBorder(channel_L_host_image, padded_L, 0, rows_L - channel_L_host_image.rows, 0, cols_L - channel_L_host_image.cols, BORDER_CONSTANT, Scalar::all(0));
Mat planes_L[] = {Mat_<float>(padded_L), Mat::zeros(padded_L.size(), CV_32F)};
Mat complexI_L;
merge(planes_L, 2, complexI_L);
dft(complexI_L, complexI_L);
// MARK: iDFT Channel_L.
Mat complexI_channel_L = complexI_L;
Mat complexI_channel_L_idft;
cv::dft(complexI_L, complexI_channel_L_idft, cv::DFT_INVERSE|cv::DFT_REAL_OUTPUT);
normalize(complexI_channel_L_idft, complexI_channel_L_idft, 0, 1, NORM_MINMAX);
imshow("complexI_channel_L_idft", complexI_channel_L_idft);
Each imshow give me different image... I think normalization would be error...
what is wrong? help!
original image
idft
OpenCV’s FFT is not normalized by default. One of the forward/backward transform pair must be normalized for the pair to reproduce the input values. Simply add cv::DFT_SCALE to the options:
cv::dft(complexI_mid_frequency_into_channel_A, iDFT_mid_frequency_into_channel_A, cv::DFT_INVERSE|cv::DFT_REAL_OUTPUT|cv::DFT_SCALE);

Is it possible to recognize so minimal changes between noisy images in OpenCV?

I want to detect the very minimal movement of a conveyor belt using image evaluation (Resolution: 31x512, image rate: 1000 per second.). The moment of belt-start is important for me.
If I do cv::absdiff between two subsequent images, I obtain very noisy result:
According to the mechanical rotation sensor of the motor, the movement starts here:
I tried to threshold the abs-diff image with a cascade of erosion and dilation, but I could detect the earliest change more than second too late in this image:
Is it possible to find the change earlier?
Here is the sequence of the Images without changes (according to motor sensor):
In this sequence the movement begins in the middle image:
Looks like I've found a solution which works in MY case.
Instead of comparing the image changes in space-domain, the cross-correlation should be applied:
I convert both images to DFT, multiply DFT-Mats and convert back. The max pixel value is the center of the correlation. As long as the images are same, the max-pix remains in the same position and moves otherwise.
The actual working code uses 3 images, 2 DFT multiplication result between images 1,2 and 2,3:
Mat img1_( 512, 32, CV_16UC1 );
Mat img2_( 512, 32, CV_16UC1 );
Mat img3_( 512, 32, CV_16UC1 );
//read the data in the images wohever you want. I read from MHD-file
//Set ROI (if required)
Mat img1 = img1_(cv::Rect(0,200,32,100));
Mat img2 = img2_(cv::Rect(0,200,32,100));
Mat img3 = img3_(cv::Rect(0,200,32,100));
//Float mats for DFT
Mat img1f;
Mat img2f;
Mat img3f;
//DFT and produtcts mats
Mat dft1,dft2,dft3,dftproduct,dftproduct2;
//Calculate DFT of both images
img1.convertTo(img1f, CV_32FC1);
cv::dft(img1f, dft1);
img2.convertTo(img3f, CV_32FC1);
cv::dft(img3f, dft3);
img3.convertTo(img2f, CV_32FC1);
cv::dft(img2f, dft2);
//Multiply DFT Mats
cv::mulSpectrums(dft1,dft2,dftproduct,true);
cv::mulSpectrums(dft2,dft3,dftproduct2,true);
//Convert back to space domain
cv::Mat result,result2;
cv::idft(dftproduct,result);
cv::idft(dftproduct2,result2);
//Not sure if required, I needed it for visualizing
cv::normalize( result, result, 0, 255, NORM_MINMAX, CV_8UC1);
cv::normalize( result2, result2, 0, 255, NORM_MINMAX, CV_8UC1);
//Find maxima positions
double dummy;
Point locdummy; Point maxLoc1; Point maxLoc2;
cv::minMaxLoc(result, &dummy, &dummy, &locdummy, &maxLoc1);
cv::minMaxLoc(result2, &dummy, &dummy, &locdummy, &maxLoc2);
//Calculate products simply fot having one value to compare
int maxlocProd1 = maxLoc1.x*maxLoc1.y;
int maxlocProd2 = maxLoc2.x*maxLoc2.y;
//Calculate absolute difference of the products. Not 0 means movement
int absPosDiff = std::abs(maxlocProd2-maxlocProd1);
if ( absPosDiff>0 )
{
std::cout << id<< std::endl;
break;
}

incorrect size of vector when converting Mat to Vector

I'm trying to convert the result Mat of templateMatch with the following code (which was found at: this question):
void convertmatVec(const cv::Mat& m, std::vector<uchar>& v) {
if (m.isContinuous()) {
v.assign(m.datastart, m.dataend);
}
else {
printf("failed to convert / not continuous");
return;
}
}
and when I check the size of the output it's not the same as the product of result's columns and rows (which is the same when I try to convert another Mat that I created):
result:
another Mat created with:
cv::Mat test(result.size(),false);
test.setTo(cv::Scalar(255));
which is then converted shows that the size is the same as the product:
So my question is how can I get the result's data so I can process it futher because I'm assuming the size of the vector should be the same as the product which it clearly isn't.
EDIT1: Added templateMatching code
void matchTemplatenoRotation(cv::Mat src, cv::Mat templ) {
cv::Mat img_display, result;
src.copyTo(img_display);
int result_cols = src.cols - templ.cols + 1;
int result_rows = src.rows - templ.rows + 1;
result.create(result_rows, result_cols, CV_32FC1);
cv::matchTemplate(src, templ, result, CV_TM_SQDIFF_NORMED);
cv::normalize(result, result, 0, 1, cv::NORM_MINMAX, -1, cv::Mat());
cv::Point minLoc, maxLoc;
double minVal, maxVal;
cv::minMaxLoc(result, &minVal, &maxVal, &minLoc, &maxLoc, cv::Mat());
cv::Point matchLoc = minLoc;
//end of templatematching
EDIT2 follow up question:
How come when I create another Mat with following code:
cv::Mat test(cv::Size(result.cols, result.rows),true);
test.setTo(cv::Scalar(255));
//cv::imshow("test3", test);
std::vector<float> testVector;
convertmatVec(test, testVector);
the vector size is as following:
You have 4 times the expected number of elements in your vector because your matrix is of type CV_32FC1. If you look at the type of m.datastart and m.dataend you will see that they are uchar* and not float* as you expect.
To correct this, change v.assign(m.datastart, m.dataend); to v.assign((float*)m.datastart, (float*)m.dataend);. And you will need to pass a vector of float instead of a vector<uchar>.
Of course, your conversion function will only work for float type matrices. you could add some tests to detect the type of the matrix inside the function.
To answer your follow up question, it appears that you have the inverse problem. You are passing a CV_8U type matrix to a function that expects a CV_32F type one. Use your old conversion function for 8-bit matrices and use the fix I suggested for 32-bit floating-values matrices. You can also add a test inside the conversion function to automatically choose the right conversion. I also advise you to read a bit on OpenCV Mat class to understand better what type of data is in your matrices.
It looks like your const cv::Mat& m has four channels.
Another thing to consider: if this Mat is the result of a matchTemplate(), then you should be using a vector<float>, instead of vector<uchar>.

Illumination normalization in OpenCV

I am working on a face recognition project. I have pictures with different lighting so I need to do illumination normalization. I read a paper which which claims to do illumination normalization. The paper describe the following function and values.
1- gamma correction with gamma = 0.2
2- Difference of Gaussian (DOG) filtering with (sigma0 = 1, sigma1 =2)
3- contrast equalization (truncation threshold of 10 and compressive component 0.1 is used in the paper)
I use CvPow for gamma correction, CvSmooth for DoG and Threshold() with truncate (I don't know how to specify the compression component) but I didn't get the exact image. I used histogram equalization for contrast equalization.
If someone has done it before or has any idea??
Link to the paper: http://lear.inrialpes.fr/pubs/2007/TT07/Tan-amfg07a.pdf
The code is below: (Python code of Peb Aryan converted to JAVACV)
public static IplImage preprocessImg(IplImage img)
{
IplImage gf = cvCreateImage(cvSize(img.width(),img.height()),IPL_DEPTH_32F, 1 );
IplImage gr = IplImage.create(img.width(),img.height(), IPL_DEPTH_8U, 1);
IplImage tr = IplImage.create(img.width(),img.height(), IPL_DEPTH_8U, 1);
IplImage b1 = IplImage.create(img.width(),img.height(),IPL_DEPTH_32F, 1 );
IplImage b2 = IplImage.create(img.width(),img.height(),IPL_DEPTH_32F, 1 );
IplImage b3 = IplImage.create(img.width(),img.height(),IPL_DEPTH_32F, 1 );
CvArr mask = IplImage.create(0,0,IPL_DEPTH_8U, 1 );
cvCvtColor(img, gr, CV_BGR2GRAY);
gamma(gr,gr,gf);
cvSmooth(gf,b1,CV_GAUSSIAN, 1);
cvSmooth(gf,b2,CV_GAUSSIAN,23);
cvSub(b1,b2,b2,mask);
cvConvertScale(b2,gr,127,127);
cvEqualizeHist(gr, gr);
//cvThreshold(gr,tr,255,0,CV_THRESH_TRUNC);
return gr;
}
public static void gamma(IplImage src,IplImage dst, IplImage temp)
{
cvConvertScale(src,temp, 1.0/255,0);
cvPow(temp, temp, 0.2);
cvConvertScale(temp, dst, 255,0);
}
Here is the result of my attempt:
And the reference from the paper:
Don't know if it's too late for you.
In the original paper, DoG was performed by a given sigma, here your radius(23) it too big. Try radius = 7 and radius = 1. About the equalization step, it's different from the paper. you need implement one by yourself.
BTW: some basic functions like cvSmooth was not implemented right for your application. You probably need to implement by yourself to get a better result.

OpenCV: How to convert CV_8UC1 mat to CV_8UC3

How to convert CV_8UC1 Mat to CV_8UC3 with OpenCV?
Mat dst;
Mat src(height, width, CV_8UC1, (unsigned char*) captureClient->data());
src.convertTo(dst, CV_8UC3);
but dst.channels() = 1
I've found that the best way to do this is:
cvtColor(src, dst, COLOR_GRAY2RGB);
The image will look the same as when it was grayscale CV_8UC1 but it will be a 3 channel image of type CV_8UC3.
From the documentation on convertTo
void Mat::convertTo(Mat& m, int rtype, double alpha=1, double beta=0) const
rtype – The desired destination matrix type, or rather, the depth (since the number of channels will be the same with the source one). If rtype is negative, the destination matrix will have the same type as the source.
You want to create a matrix for each of the 3 channels you want to create and then use the merge function. See the answers to this question
The convention is, that for the type CV_8UC3, the pixels values range from 0 to 255, and for type CV_32FC3 from 0.0 to 1.0. Thus you need to use a scaling factor of 255.0, instead of 1.0:
Mat::convertTo(newImage, CV_32FC1, 255.0);

Resources