Related
In my application the JPEG image arrives as a char* array. I need to convert this to an OpenCV cv::Mat object. To test behavior I create another cv::Mat from cv::imread.
The Mat create from imread works fine when I try to convert it from BGR to RGB format. But the Mat created from the raw char array Seg faults when trying to convert it.
Code is attached below:
//read_image -> get the char* array
//convert_image -> convert char* array to Mat object
//mat_process -> convert the Matrix to the RGB format of the correct size
//When the Matrix is created from cv::imread and I call mat_process it returns correctly
//When the Matrix is created from read_image -> convert_image, it seg faults on mat_process
typedef cv::Point3_<float> Pixel;
char* read_image(string image_name){
cout << "Reading image " << endl;
std::ifstream file(image_name, ios::binary);
file.seekg(0, std::ios::end);
size_t filesize = (int)file.tellg();
file.seekg(0);
char* output = new char[filesize];
file.read(output, filesize);
cout << "IMAGE SIZE IS " << filesize << endl;
return output;
}
cv::Mat convert_image(char* img){
cv::Mat* raw_data= new cv::Mat(1080,1920,CV_8UC3,img,cv::Mat::AUTO_STEP);
cout << " Got raw data " << raw_data->cols << " rows " << raw_data->rows << endl;
//Mat decoded_mat = imdecode(raw_data,IMREAD_UNCHANGED);
//cout << "Decoded the matrix " << decoded_mat.cols << " rows are " << decoded_mat.rows << endl;
//return decoded_mat;
return *raw_data;
}
auto mat_process(cv::Mat src, uint width, uint height) -> cv::Mat{
// convert to float; BGR -> RGB
cv::Mat dst;
cout << "Creating dst" << endl;
src.convertTo(dst, CV_32FC3); **// SEGMENTATION FAULT OCCURS HERE WITH RAW CHAR ARRAY**
cout << "Creating dst2" << endl;
cv::cvtColor(dst, dst, cv::COLOR_BGR2RGB);
cout << "Creating dst3" << endl;
// normalize to -1 & 1
Pixel* pixel = dst.ptr<Pixel>(0,0);
cout << "Creating dst4" << endl;
const Pixel* endPixel = pixel + dst.cols * dst.rows;
cout << "Creating dst5" << endl;
for (; pixel != endPixel; pixel++)
normalize(*pixel);
// resize image as model input
cout << "Creating dst6" << endl;
cv::resize(dst, dst, cv::Size(width, height));
cout << "Creating dst7" << endl;
return dst;
}
char * img1 = read_image(imgf1);
cv::Mat img1_mat = convert_image(img1);
cv::Mat img1_comp = imread(imgf1);
cout << " Start process " <<endl;
cv::Mat conv1 = mat_process(img1_comp,320,320); //This code executes fine
cout << " Done 1 " << endl;
cv::Mat conv2 = mat_process(img1_mat,320,320); //Segmentation Fault appears here
cout << " Done process " <<endl;
I have encountered on designing program to allow capturing images every second from video files (avi, mp4, etc...).
First, I was able to capture images frame by frame from video file.
Second, I was able to analyze pixel color values from images in the same folder at the same time and saved pixel value in the txt file.
And here I have some problem. I am now trying to combine these two codes at once, but I have strange results. I refer the code below.
int main(){
VideoCapture cap("D:\\data\\extra\\video200ul.avi");
if (!cap.isOpened())
return -1;
Ptr<BackgroundSubtractor> pMOG2 = createBackgroundSubtractorMOG2(20, 16, true);
Mat fg_mask;
Mat frame;
int count = 0;
String name, folder;
for (;;) {
// Get frame
cap >> frame; // get a new frame from video
++count;
// Update counter
// Background subtraction
if (count % 2 == 0) {
pMOG2->apply(frame, fg_mask, 0.001);
cout << count << endl;
if (!frame.empty()) {
imshow("frame", frame);
// imshow("fg_mask", fg_mask);
}
// Save foreground mask
name = "mask" + std::to_string(count) + ".png";
// string name = "mask_" + std::to_string(static_cast<long long>(count) + ".png";
folder = imwrite("D:\\data\\extra\\" + name, frame);
}
anal(folder);
}
waitKey(0);
return 0;
}
First, The code above I wrote is for capturing images frame by frame from video file. However, if I got the images per frame, I will have so many pictures on my folder, so I would like to capture an image per second from the video file. I have tried to use CV_CAP_PROP_POS_MSEC instead using cap << frame, but it did not work for me.
Second, when I merge this code to another code what I wrote below, it showed some error messages like, "libpng warning image width, length, data are zero in ihdr."
int anal(String folder) {
folder = "D:\\data\\extra\\*.png";
vector<String> filenames;
glob(folder, filenames);
cv::Mat ori_image;
for (size_t i = 0; i < filenames.size(); ++i) {
ori_image = imread(filenames[i], IMREAD_COLOR);
if (ori_image.empty()) {
cout << "Check your file again." << std::endl;
return -1;
}
rectangle(ori_image, Point(215, 98), Point(245, 110), Scalar(0, 255, 255), 1);
imshow("Original Image", ori_image);
cv::Scalar sums;
sums = cv::sum(ori_image);
double totalSum = sums[0] + sums[1] + sums[2];
if (totalSum <= 0) {
cout << "$$ RGB percentage $$" << " \n\n";
cout << "R: " << 100.0 / 3 << " % \n";
cout << "G: " << 100.0 / 3 << " % \n";
cout << "B: " << 100.0 / 3 << " % \n\n";
}
else {
cout << "$$ RGB percentage $$" << " \n\n"; // red value
cout << "R: " << sums[2] / totalSum * 100 << " % \n"; // red value
cout << "G: " << sums[1] / totalSum * 100 << " % \n"; // green value
cout << "B: " << sums[0] / totalSum * 100 << " % \n\n"; // blue value
}
}
as I prepared the code above, I tried to calculate red, blue, green percentages of all the captured images from the video. However, when I separate these two code and run them, they worked fine, but if I merge them together, It showed error messages.
I would like to combine these two code for analysis for color values from the captured images at video every second.
Please help me out this problem.
Thank you in advance.
-----------Edited part----------------------
I used your revised version and applied to my updated code,
void imageAnalysis(std::string folder, cv::Mat frame){
cv::Mat ori_image = frame.clone();
std::string path = folder;
cv::rectangle(ori_image, Point(215, 105), Point(245, 120), Scalar(0, 255, 255), 1);
cv::imshow("Original Image", ori_image);
cv::waitKey(1);
String folder = "D:\\data\\dfdf\\*.png";
vector<String> filenames;
cv::glob(path, filenames);
for (size_t t = 0; t < filenames.size(); t++) {
ori_image = imread(filenames[t], IMREAD_COLOR); // ori_image
if (ori_image.empty()) { //ori_image
cout << "Check your file again." << "\n";
break;
//return -1;
}
rectangle(ori_image, Point(215, 105), Point(245, 120), Scalar(0, 255, 255), 1);
imshow("Original Image", ori_image);
cv::waitKey(1);
Mat image_HSV;
cvtColor(ori_image, image_HSV, CV_BGR2HSV);
double h = 0.0;
double s = 0.0;
double v = 0.0;
int col = image_HSV.cols; // 480
int row = image_HSV.rows; // 272
int corow = ((col - 235) - 215) * ((row - 152) - 108);
Mat mask;
inRange(image_HSV, Scalar(100, 0, 0), Scalar(100, 255, 255), mask); // convert binary
image_HSV.setTo(Scalar(0, 0, 0), mask);
for (int i = 108; i < row - 152; i++) {
for (int j = 215; j < col - 235; j++) {
Vec3b hsv = image_HSV.at<cv::Vec3b>(i, j);
h += (int)(hsv.val[0]);
s += (int)(hsv.val[1]);
v += (int)(hsv.val[2]);
if (hsv[0] != 100) {
hsv[0] = 0;
hsv[1] = 0;
hsv[2] = 0;
}
}
}
cout << "$$ Hue(H), Saturation(S), Brightness(V) $$" << filenames[t] << " !! \n\n";
cout << "H: " << h / corow * 360 / 180 << " % \n"; //
cout << "S: " << s / corow * 100 / 255 << " % \n";
cout << "V: " << v / corow * 100 / 255 << " % \n\n";
std::ofstream file("D:\\data\\dfdf\\result_4.txt", std::ios_base::app);
file << v / corow * 100 / 255 << " \n"; // v value
file.close();
}
}
As you can see the imageAnalysis() function, I added std::string folder for the path of extracted images from video clip. However, when I applied this code, I have really weird results like below..
enter image description here
I thought I am supposed to get color value from every 24th image but as you see the results above, I got color values from all images in random order.
Thank you in advance.
It was really nice to learn how to code in efficient way!!
Just to clear the error you mentioned about CV_CAP_PROP_POS_MSEC in your comments:
when I apply CV_CAP_PROP_POS_MSEC to my code, I found some error
messages like "CV_CAP_PROP_POS_MSEC is not defined."
A lot of the constant values are scoped in OpenCV. That means, CV_CAP_PROP_POS_MSEC is not defined, but cv::CV_CAP_PROP_POS_MSEC is. You can also obtain the FPS with cv::CAP_PROP_FPS.
Now to your code, I would actually do something that does not require to save and load the image, but rather pass the images to be processed, like this:
#include "opencv2/opencv.hpp"
#include <iostream>
int main(){
cv::VideoCapture cap("D:\\data\\extra\\video200ul.avi");
if (!cap.isOpened())
{
std::cout << "Could not open video" << std::endl;
return -1;
}
cv::Ptr<cv::BackgroundSubtractor> pMOG2 = cv::createBackgroundSubtractorMOG2(20, 16, true);
cv::Mat fg_mask, frame;
int count = 0;
const int fps = 24; // you may set here the fps or get them from the video
std::string name, folder;
// with cap.read you can check already if the video ended
while (cap.read(frame)) {
// Background subtraction
if (count % fps == 0) {
pMOG2->apply(frame, fg_mask, 0.001);
// Save foreground mask
name = "mask" + std::to_string(count) + ".png";
bool result = cv::imwrite("D:\\data\\extra\\" + name, frame);
imageAnalysis(frame, count);
}
// at the end of the loop so that the first image is used
++count;
}
cv::waitKey(0);
return 0;
}
And the imageAnalysis function is defined as:
// You can pass cv::Mat as value, it is almost like a smart pointer
void imageAnalysis(cv::Mat frame, int count)
{
cv::Mat ori_image = frame.clone();
cv::rectangle(ori_image, Point(215, 98), Point(245, 110), Scalar(0, 255, 255), 1);
// each imshow needs a waitKey to update the window in which it is being shown
cv::imshow("Original Image", ori_image);
cv::waitKey(1);
cv::Scalar sums;
sums = cv::sum(ori_image);
double totalSum = sums[0] + sums[1] + sums[2];
std::ofstream output("D:\\data\\extra\\mask" + std::to_string(count) + ".txt");
if (totalSum <= 0)
{
std::cout << "$$ RGB percentage $$" << std::endl << std::endl;
std::cout << "R: " << 100.0 / 3 << std::endl;
std::cout << "G: " << 100.0 / 3 << std::endl;
std::cout << "B: " << 100.0 / 3 << std::endl << std::endl;
output << "$$ RGB percentage $$" << std::endl << std::endl;
output << "R: " << 100.0 / 3 << std::endl;
output << "G: " << 100.0 / 3 << std::endl;
output << "B: " << 100.0 / 3 << std::endl << std::endl;
}
else {
std::cout << "$$ RGB percentage $$" << std::endl << std::endl;
std::cout << "R: " << sums[2] / totalSum * 100 << std::endl; // red value
std::cout << "G: " << sums[1] / totalSum * 100 << std::endl; // green value
std::cout << "B: " << sums[0] / totalSum * 100 << std::endl << std::endl; // blue value
output << "$$ RGB percentage $$" << std::endl << std::endl;
output << "R: " << sums[2] / totalSum * 100 << std::endl; // red value
output << "G: " << sums[1] / totalSum * 100 << std::endl; // green value
output << "B: " << sums[0] / totalSum * 100 << std::endl << std::endl; // blue value
}
}
Some comments of the code above, I replaced the cap >> frame to cap.read(frame). It is the same functionality, but the later gives a bool result that is false if it could not grab the image, like if the video is over. I change the count add at the end, yo you get the frames 0,23,... this way the first one will be use as well. Finally, you should use the namespaces cv::, std:: etc. This is just best practice, it avoids ambiguities and problems that may arise with certain libraries.
If you do not need the image in disk, but only the analysis, then remove the saving part and pass every frame to the imageAnalysis function, this way you may have more data for your statistics. Also, consider returning the cv:Scalar of sums in the function and then you can do some statistics of the whole second or the whole video.
If you have any question, feel free to ask in the comments.
I need to load image of any possible supported type and convert it to BGRA (CV_8UC4).
#include <opencv2/opencv.hpp>
#include <iostream>
int main()
{
std::string filePath;
cv::Mat image = cv::imread(filePath, CV_LOAD_IMAGE_UNCHANGED);
std::cout << "HEIGHT: " << image.rows << " Width: " << image.cols << std::endl;
std::cout << "CHANNELS: " << image.channels() << std::endl;
switch(image.channels())
{
case 1:
{
cvtColor(image, image, CV_GRAY2BGRA);
std::cout << "CHANNELS: " << image.channels() << std::endl;
break;
}
case 3:
{
cvtColor(image, image, CV_BGR2BGRA);
std::cout << "CHANNELS: " << image.channels() << std::endl;
break;
}
}
switch(image.depth())
{
case CV_8U:
{
/* DO NOTHING AS IT IS DESIRED */
break;
}
default:
{
image.convertTo(image, CV_8U);
image.depth() == CV_8U ? std::cout << "CONVERSION SUCCESSFUL" << std::endl : std::cout << "CONVERSION UNSUCCESSFUL" << std::endl;
}
}
cv::resize(image, image, cvSize(1920, 1080));
/*************Now do desired operations*******************/
}
Just want to make sure if there is no condition missing. Can you please check if for some type of image this may fail to give me CV_8UC4 image??
Please suggest if there is any better approach possible.
******Edit*******
I was able to achieve the above. But now I am facing another problem
I need to operate on images with size less than 1920x1080 so when I load the image, I resize it to smaller size.
The problem I am facing is with images of very large resolution (since I can use just 500 mb of memory but the allocation size of cv::Mat is exceeding that. Therefore even when I have to operate on scaled image (smaller resolution) I cannot load the original image before resizing.
Is there an option to scale the image while loading so that cv::Mat uses memory according to the scaled version of the image?
Thanks for the help.
I don't understand why output is a strange number when I run my code:
int main(int argc, char** argv)
{
Mat im;
im = imread("lena.png", CV_LOAD_IMAGE_GRAYSCALE);
cout << im.at<uchar>(0, 0) << endl;
waitKey(0);
}
If I visualize image I see the correct image.
Where am I wrong?
Because it shows the symbol, like cout << char(123) << endl;
You have to use int cast:
cout << (int) im.at<uchar>(0, 0) << endl;
As stated in the official documentation you don´t get the actual intensity value directly but a scalar.
Try this:
Scalar intensity = im.at<uchar>(0, 0);
cout << intensity.val[0] << endl;
and for images with more than one channel you can use:
Vec3b intensity = im.at<Vec3b>(0, 0);
cout << intensity.val[0] << intensity.val[1] << intensity.val[2] << endl;
hi, i'm trying to play a little bit with Mat class.
I want to do a product element wise between two images, the c++/opencv port of MATLAB immultiply.
This is my code:
#include <opencv2/core/core.hpp>
#include <opencv2/highgui/highgui.hpp>
#include "opencv2/imgproc/imgproc.hpp"
#include <iostream>
using namespace cv;
using namespace std;
Mat imgA, imgB;
Mat imgAB;
Mat product;
void printMinMax(Mat m, string s) {
double minVal;
double maxVal;
Point minLoc;
Point maxLoc;
minMaxLoc( m, &minVal, &maxVal, &minLoc, &maxLoc );
cout << "min val in " << s << ": " << minVal << endl;
cout << "max val in " << s << ": " << maxVal << endl;
}
int main(int /*argc*/, char** /*argv*/) {
cout << "OpenCV version: " << CV_MAJOR_VERSION << " " << CV_MINOR_VERSION << endl;
imgA = imread("test1.jpg");
cout << "original image size: " << imgA.rows << " " << imgA.cols << endl;
cout << "original type: " << imgA.type() << endl;
cvtColor(imgA, imgA, CV_BGR2GRAY);
printMinMax(imgA, "imgA");
imgB = imread("test2.jpg");
cout << "original image size: " << imgB.rows << " " << imgB.cols << endl;
cout << "original type: " << imgB.type() << endl;
cvtColor(imgB, imgB, CV_BGR2GRAY);
printMinMax(imgB, "imgB");
namedWindow("originals", CV_WINDOW_AUTOSIZE);
namedWindow("product", CV_WINDOW_AUTOSIZE);
imgAB = Mat( max(imgA.rows,imgB.rows), imgA.cols+imgB.cols, imgA.type());
imgA.copyTo(imgAB(Rect(0, 0, imgA.cols, imgA.rows)));
imgB.copyTo(imgAB(Rect(imgA.cols, 0, imgB.cols, imgB.rows)));
product = imgA.mul(imgB);
printMinMax(product, "product");
while( true )
{
char c = (char)waitKey(10);
if( c == 27 )
{ break; }
imshow( "originals", imgAB );
imshow( "product", product );
}
return 0;
}
here is the result:
OpenCV version: 2 4
original image size: 500 500
original type: 16
min val in imgA: 99
max val in imgA: 255
original image size: 500 500
original type: 16
min val in imgB: 0
max val in imgB: 255
init done
opengl support available
min val in product: 0
max val in product: 255
I think that max value in the product has to be greater than 255, but is truncated to 255 because the type of the two matrixes is 16.
I have tried to convert the matrixes to CV_32F but the maxVal in the product is 64009 (a number that i don't understand)
Thanks to Wajih comment i have done some basic test, and some basic debug, and i got i work perfectly. I think this could become a mini tutorial on alpha blending and image multiply, but for now is only a few lines of commented code.
note that the 2 images must be of the same size.. and for sure some error checking should be done for a solid code..
Hope it helps someone! And, of course, if you have some hints to make this code more readable or more compact (one-liner guys are very appreciate!) or efficient.. just comment, thank you a lot!
#include <opencv2/core/core.hpp>
#include <opencv2/highgui/highgui.hpp>
#include "opencv2/imgproc/imgproc.hpp"
#include <iostream>
using namespace cv;
using namespace std;
void printMinMax(Mat m, string name) {
double minVal;
double maxVal;
Point minLoc;
Point maxLoc;
if(m.channels() >1) {
cout << "ERROR: matrix "<<name<<" must have 1 channel for calling minMaxLoc" << endl;
}
minMaxLoc( m, &minVal, &maxVal, &minLoc, &maxLoc );
cout << "min val in " << name << ": " << minVal << " in loc: " << minLoc << endl;
cout << "max val in " << name << ": " << maxVal << " in loc: " << maxLoc << endl;
}
int main(int /*argc*/, char** /*argv*/) {
cout << "OpenCV version: " << CV_MAJOR_VERSION << " " << CV_MINOR_VERSION << endl; // 2 4
Mat imgA, imgB;
Mat imgAB;
Mat product;
// fast matrix creation, comma-separated initializer
// example1: create a matrix with value from 0 to 255
imgA = Mat(3, 3, CV_8UC1);
imgA = (Mat_<uchar>(3,3) << 0,1,2,3,4,5,6,7,255);
cout << "test Mat 3x3" << endl << imgA << endl;
// not that if a value exceed 255 it is truncated at value%256
imgA = (Mat_<uchar>(3,3) << 0,1, 258 ,3,4,5,6,7,255);
cout << "test Mat 3x3 with last element truncated to 258%256=2" << endl << imgA << endl;
// create a second matrix
imgB = Mat(3, 3, CV_8UC1);
imgB = (Mat_<uchar>(3,3) << 0,1,2,3,4,5,6,7,8);
// now the matrix product. we are multiplying a value that can goes from 0-255 with another 0-255 value..
// the edge cases are "min * min" and "max * max",
// that means: our product is a function that return a value in the domain 0*0-255*255 ; 0-65025
// ah, ah! this number exceed the Mat U8C1 domain!, we need different data types.
// we need a bigger one.. let's say 32FC1
Mat imgA_32FC1 = imgA.clone();
imgA_32FC1.convertTo(imgA_32FC1, CV_32FC1);
Mat imgB_32FC1 = imgB.clone();
imgB_32FC1.convertTo(imgB_32FC1, CV_32FC1);
// after conversion.. value are scaled?
cout << "imgA after conversion:" << endl << imgA_32FC1 << endl;
cout << "imgB after conversion:" << endl << imgB_32FC1 << endl;
product = imgA_32FC1.mul( imgB_32FC1 );
// note: the product values are in the range 0-65025
cout << "the product:" << endl << product << endl;
// now, this does not have much sense, because we started from a 0-255 range Mat and now we have a 0-65025 that is nothing..
// it is not uchar range and it is not float range (that is a lot bigger than that)
// so, we can normalize back to 0-255
// what do i mean with 'normalize' now?
// i mean: scale all values for a constant that maps 0 to 0 and 65025 to 255..
product.convertTo(product, CV_32FC1, 1.0f/65025.0f * 255);
// but it is still a 32FC1.. not as the start matix..
cout << "the product, normalized back to 0-255, still in 32FC1:" << endl << product << endl;
product.convertTo(product, CV_8UC1);
cout << "the product, normalized back to 0-255, now int 8UC1:" << endl << product << endl;
cout << "-----------------------------------------------------------" << endl;
// real stuffs now.
imgA = imread("test1.jpg");
cvtColor(imgA, imgA, CV_BGR2GRAY);
imgB = imread("test2.jpg");
cvtColor(imgB, imgB, CV_BGR2GRAY);
imgA_32FC1 = imgA.clone();
imgA_32FC1.convertTo(imgA_32FC1, CV_32FC1);
imgB_32FC1 = imgB.clone();
imgB_32FC1.convertTo(imgB_32FC1, CV_32FC1);
product = imgA_32FC1.mul( imgB_32FC1 );
printMinMax(product, "product");
product.convertTo(product, CV_32FC1, 1.0f/65025.0f * 255);
product.convertTo(product, CV_8UC1);
// concat two images in one big image
imgAB = Mat( max(imgA.rows,imgB.rows), imgA.cols+imgB.cols, imgA.type());
imgA.copyTo(imgAB(Rect(0, 0, imgA.cols, imgA.rows)));
imgB.copyTo(imgAB(Rect(imgA.cols, 0, imgB.cols, imgB.rows)));
namedWindow("originals", CV_WINDOW_AUTOSIZE);
namedWindow("product", CV_WINDOW_AUTOSIZE);
while( true )
{
char c = (char)waitKey(10);
if( c == 27 )
{ break; }
imshow( "originals", imgAB );
imshow( "product", product );
}
return 0;
}
You are right, you should convert your matrices imgA, imgB to say CV32FC1 type. Since the max values in this matrices is 255, the maximum possible value is 65025. However, the maximum at imgA and imgB may not be in the same location, so 64009 is quite possible.