Introduction
Digital Image Processing
This page contains codes and exercises that were developed for the Digital Image Processing course (2018.2) at the Universidade Federal do Rio Grande do Norte (UFRN - Brazil).
Unit 1
In this unit, the following examples were developed:
- Negative of a region
- Swapping regions
- Counting bubbles with holes
- Histogram equalization
- Motion detection
- Laplacian of Gaussian
- Tilty Shift
Exercise 1 - Negative of a region
This is the first exercise using OpenCV image manipulation. In this example, we'll be reading and processing a region of a given image in order to create a negative effect on it. To achive this goal, the following fomula is used to invert the color of the pixels: pixel_color = 255 - pixel_color
.
As the image used is in grayscale, we will be aplying it for one channel only. The region is seleted by a couple of for-loops that go through an area within the image, inverting its pixels, as shown below.
#include <iostream>
#include <opencv2/opencv.hpp>
using namespace cv;
using namespace std;
int main()
{
Mat image;
int pxl;
image= imread("img/camera__.jpg",CV_LOAD_IMAGE_GRAYSCALE);
if(!image.data)
cout << "couldn't load image" << endl;
namedWindow("window",WINDOW_AUTOSIZE);
for(int i=100;i<200;i++){
for(int j=100;j<200;j++){
pxl = image.at<uchar>(i,j);
image.at<uchar>(i,j) = 255 - pxl;
}
}
imwrite("negative.jpg", image);
imshow("window", image);
waitKey();
return 0;
}
Exercise 2 - Swapping regions
In this example, we'll create an image by swapping regions from an original one. For this, Regions of Interest (ROI) were created using rectangles with half the height and width of the original image. In sequence, using the Mat (const Mat &m, const Rect &roi)
constructor, 4 images were created by using those rectangles and finally the CV::Mat::copyTo(OutPutArray m)
method was used to perfom the swap of the 1-3 and 2-4 regions.
#include <iostream>
#include <opencv2/opencv.hpp>
using namespace cv;
using namespace std;
int main()
{
Mat image = imread("img/camera__.jpg",CV_LOAD_IMAGE_GRAYSCALE);
if(!image.data)
cout << "nao abriu camera__.jpg" << endl;
namedWindow("janela",WINDOW_AUTOSIZE);
imshow("janela", image);
waitKey();
int x = 0, y = 0, width = image.rows/2, height = image.cols/2;
Mat image2;
image.copyTo(image2);
// creating Regions of Interest
Rect rois[4] = {
Rect( 0, 0, width, height),
Rect(width, 0, width, height),
Rect( 0, height, width, height),
Rect(width, height, width, height)
};
// copying regions and swapping them
Mat region1 (image, rois[3]);
region1.copyTo(image2(rois[0]));
Mat region2 (image, rois[2]);
region2.copyTo(image2(rois[1]));
Mat region3 (image, rois[1]);
region3.copyTo(image2(rois[2]));
Mat region4 (image, rois[0]);
region4.copyTo(image2(rois[3]));
imwrite("swapping.jpg", image2);
imshow("janela", image2);
waitKey();
return 0;
}
Exercise 3. Labeling objects
Example: Bubbles with holes
In this example, we'll introduce the labeling concept, a method which allow us to identify objects or regions in an image. To create labels, we made use of the floodfill
OpenCV method, which consists of painting an area that has the same color, replacing it by another color (label). Here this technique is used for labeling and counting white bubbles in a black background,identifing and counting as well, bubbles with holes.
This example was based on the labeling.cpp algorithm and answers to the proposed exercises at agostinhobritojr.github.io.
The labeling.cpp algorithm works very well in cases where the amount of objects are below 255 units. But, if we have more than 255 objects to label, varying from the pixel grayscale intesity, we'll find an issue. For theses cases, another approach could be used, for instance, a diferent number representation as such as floating point can be used, so a much greater amount of objects could be labeled.
#include <iostream>
#include <opencv2/opencv.hpp>
using namespace cv;
int main(){
Mat image, mask;
int width, height;
int nobjects, holes, counter;
// - Reading image --
CvPoint p;
image = imread("img/bolhas.png",CV_LOAD_IMAGE_GRAYSCALE);
if(!image.data){
std::cout << "Couldn't load the file correctly\n";
return(-1);
}
width=image.size().width;
height=image.size().height;
p.x = 0;
p.y = 0;
// -- Excluding objects that touch the first and last lines --
for(int i=0; i<height; i++){
if(image.at<uchar>(0, i) == 255){
floodFill(image,Point(i, 0), 0);
}
if(image.at<uchar>(width-1, i) == 255){
floodFill(image,Point(i, width-1), 0);
}
}
// -- Excluding objects that touch the first and last columns --
for(int i=0; i<width; i++){
if(image.at<uchar>(i,0) == 255){
floodFill(image,Point(0, i), 0);
}
if(image.at<uchar>(i,height-1) == 255){
floodFill(image,Point(height-1, i), 0);
}
}
// -- Looking for objects and labeling them --
nobjects=0;
for(int i=0; i<height; i++){
for(int j=0; j<width; j++){
if(image.at<uchar>(i,j) == 255){
// Found an white object
nobjects++;
p.x=j;
p.y=i;
floodFill(image,p,nobjects);
}
}
}
// -- Changing background Color --
floodFill(image,Point(0,0),255);
// -- Looking for bubbles with holes --
holes = 0; counter = 1;
for(int i=0; i<height; i++){
for(int j=0; j<width; j++){
if(image.at<uchar>(i,j) == 0 && (int)image.at<uchar>(i,j-1)>counter){ //se encontrar um buraco e já não tiver contado a bolha
// Found a bubble with a hole
counter++;
p.x=j-1;
p.y=i;
floodFill(image,p,counter);
}
}
}
std:: cout << "Number of bubbles: " << nobjects << " and number of bubbles with holes: " << counter <<std::endl;
imwrite("labeling.png", image);
imshow("janela", image);
waitKey();
return 0;
}
As a result of the image processing, the output is:
Exercise 4. Equalizing an image
An image histogram is a graphical representation of the intensity distribution of an image. It quantifies the number of pixels for each intensity value considered. OpenCV - Histogram equalization Histogram equalization is a method that improves the contrast in an image, in order to stretch out the intensity range. In this example, in equalize.cpp this outcome has been achived by using OpenCV function:equalize_hist:equalizeHist <>.
This example was based on the histogram.cpp algorithm and answers to the proposed exercises at agostinhobritojr.github.io.
#include <iostream>
#include <opencv2/opencv.hpp>
using namespace cv;
using namespace std;
/** @function main */
int main( int argc, char** argv )
{
Mat src, dst;
string source_window = "Source image";
string equalized_window = "Equalized Image";
/// Camera
VideoCapture cap;
//start camera
cap.open(0);
if(!cap.isOpened()){
cout << "Camera not available";
return -1;
}
while(1){
/// capture image
cap >> src;
/// Convert to grayscale
cvtColor( src, src, CV_BGR2GRAY );
/// Apply Histogram Equalization
equalizeHist( src, dst );
/// Display results
namedWindow( source_window, CV_WINDOW_NORMAL );
namedWindow( equalized_window, CV_WINDOW_NORMAL );
imshow( source_window, src );
imshow( equalized_window, dst );
if(waitKey(30) >= 0) break;
}
// Closes all the windows
destroyAllWindows();
return 0;
}
What does this program do?
Create a VideoCapture object
Capture frame-by-frame
Convert the original image to grayscale
Equalize the Histogram by using the OpenCV function EqualizeHist
Display the source and equalized images in a window.
As we can see from the output image, the equalized image has a better distribution of the overall contrast, the furniture behind can be easily observed now. Original image is on the rigth and equalized on the left side
Exercise 5. Motion Detection
Analizing an image histogram, it is possible to observe several aspects of an image or a video, such as motion detection. In this example, we use the correlation between two frames of an video to determine if there was motion or not in it.
To achive this goal, the OpenCV CalcHist
function was used to get the histogram of the frame, and the comparison between two frames was perfomed by the compareHist
function, through the correlation comparison.;
What does this program do?
Create a VideoCapture object
Capture frame-by-frame
Calculate histogram by using the OpenCV function
Display the source and equalized images in a window.
Calculates a histogram of a set of arrays:
C++: void calcHist(const Mat* images, int nimages, const int* channels, InputArray mask, OutputArray hist, int dims, const int* histSize, const float** ranges, bool uniform=true, bool accumulate=false )
OpenCV implements the function compareHist to perform a comparison. It also offers 4 different metrics to compute the matching, in this program we used the correlation as an metric.
compareHist(histR, old_hist, CV_COMP_CORREL);
This example was based on the histogram.cpp algorithm and answers to the proposed exercises at agostinhobritojr.github.io.
#include <iostream>
#include <opencv2/opencv.hpp>
using namespace cv;
using namespace std;
int main(int argc, char** argv){
Mat image;
int width, height;
VideoCapture cap;
vector<Mat> planes;
Mat histR, motion, old_hist;
int nbins = 64;
float range[] = {0, 256};
const float *histrange = { range };
bool uniform = true;
bool acummulate = false;
// Start Camera
cap.open(0);
if(!cap.isOpened()){
cout << "cameras indisponiveis";
return -1;
}
// image size
width = cap.get(CV_CAP_PROP_FRAME_WIDTH);
height = cap.get(CV_CAP_PROP_FRAME_HEIGHT);
cout << "largura = " << width << endl;
cout << "altura = " << height << endl;
// mat for histogram - analyzing only red here.
int histw = nbins, histh = nbins/2;
Mat histImgR(histh, histw, CV_8UC3, Scalar(0,0,0));
// Motion alert
motion = imread("img/motion_detected.png",CV_LOAD_IMAGE_COLOR);
if(!motion.data)
cout << "couldn't load image" << endl;
namedWindow("window",WINDOW_AUTOSIZE);
// faz uma vez para poder, calcular a primeira correlação, dar continuidade dentro do loop
cap >> image;
split (image, planes);
calcHist(&planes[0], 1, 0, Mat(), histR, 1,
&nbins, &histrange,
uniform, acummulate);
normalize(histR, histR, 0, histImgR.rows, NORM_MINMAX, -1, Mat());
histR.copyTo(old_hist);
while(1){
histR.copyTo(old_hist);
cap >> image;
split (image, planes);
calcHist(&planes[0], 1, 0, Mat(), histR, 1,
&nbins, &histrange,
uniform, acummulate);
normalize(histR, histR, 0, histImgR.rows, NORM_MINMAX, -1, Mat());
histImgR.setTo(Scalar(0));
double compare = compareHist(histR, old_hist, CV_COMP_CORREL);
cout << compare << endl;
if ( compare < 0.01)
{
cout << "Motion detected\n";
motion.copyTo(image(Rect(0, histh, motion.size().width, motion.size().height)));
}
for(int i=0; i<nbins; i++){
line(histImgR,
Point(i, histh),
Point(i, histh-cvRound(histR.at<float>(i))),
Scalar(0, 0, 255), 1, 8, 0);
}
histImgR.copyTo(image(Rect(0, 0 ,nbins, histh)));
imshow("image", image);
if(waitKey(30) >= 0) break;
old_hist.copyTo(histR);
}
return 0;
}
Exercise 6. Laplacian of a Gaussian filter
This example was based on the filtroespacial.cpp algorithm and answers to the proposed exercises at agostinhobritojr.github.io.
This program adds one more function to the provided example. Through this new function, the laplacian of the gaussian of a captured image can be calculated by pressing the key "z" and the filtered image is displayed.
Comparing both laplacian and laplacion of gauss filters results, it is possible to conclude that the second filter returns a better filtering, as we can observe more easily the borders and cornes of the image.
#include <iostream>
#include <opencv2/opencv.hpp>
using namespace cv;
using namespace std;
void printmask(Mat &m){
for(int i=0; i<m.size().height; i++){
for(int j=0; j<m.size().width; j++){
cout << m.at<float>(i,j) << ",";
}
cout << endl;
}
}
void menu(){
cout << "\npressione a tecla para ativar o filtro: \n"
"a - calcular modulo\n"
"m - media\n"
"g - gauss\n"
"v - vertical\n"
"h - horizontal\n"
"l - laplaciano\n"
"z - laplaciano do gaussiano\n"
"esc - sair\n";
}
int main(int argvc, char** argv){
VideoCapture video;
float media[] = {1,1,1,
1,1,1,
1,1,1};
float gauss[] = {1,2,1,
2,4,2,
1,2,1};
float horizontal[]={-1,0,1,
-2,0,2,
-1,0,1};
float vertical[]={-1,-2,-1,
0,0,0,
1,2,1};
float laplacian[]={0,-1,0,
-1,4,-1,
0,-1,0};
float lapgauss[]={0,0,1,0,0,
0,1,2,1,0,
1,2,-16,2,1,
0,1,2,1,0,
0,0,1,0,0};
Mat cap, frame, frame32f, frameFiltered;
Mat mask(3,3,CV_32F), mask1;
Mat result, result1;
double width, height, min, max;
int absolut;
char key;
video.open(0);
if(!video.isOpened())
return -1;
width=video.get(CV_CAP_PROP_FRAME_WIDTH);
height=video.get(CV_CAP_PROP_FRAME_HEIGHT);
std::cout << "largura=" << width << "\n";;
std::cout << "altura =" << height<< "\n";;
namedWindow("filtroespacial",1);
mask = Mat(3, 3, CV_32F, media);
scaleAdd(mask, 1/9.0, Mat::zeros(3,3,CV_32F), mask1);
swap(mask, mask1);
absolut=1; // calcs abs of the image
menu();
for(;;){
video >> cap;
cvtColor(cap, frame, CV_BGR2GRAY);
flip(frame, frame, 1);
imshow("original", frame);
frame.convertTo(frame32f, CV_32F);
filter2D(frame32f, frameFiltered, frame32f.depth(), mask, Point(1,1), 0);
if(absolut){
frameFiltered=abs(frameFiltered);
}
frameFiltered.convertTo(result, CV_8U);
imshow("filtroespacial", result);
key = (char) waitKey(10);
if( key == 27 ) break; // esc pressed!
switch(key){
case 'a':
menu();
absolut=!absolut;
break;
case 'm':
menu();
mask = Mat(3, 3, CV_32F, media);
scaleAdd(mask, 1/9.0, Mat::zeros(3,3,CV_32F), mask1);
mask = mask1;
printmask(mask);
break;
case 'g':
menu();
mask = Mat(3, 3, CV_32F, gauss);
scaleAdd(mask, 1/16.0, Mat::zeros(3,3,CV_32F), mask1);
mask = mask1;
printmask(mask);
break;
case 'h':
menu();
mask = Mat(3, 3, CV_32F, horizontal);
printmask(mask);
break;
case 'v':
menu();
mask = Mat(3, 3, CV_32F, vertical);
printmask(mask);
break;
case 'l':
menu();
mask = Mat(3, 3, CV_32F, laplacian);
printmask(mask);
break;
case 'z':
menu();
mask = Mat(5, 5, CV_32F, lapgauss);
printmask(mask);
default:
break;
}
}
return 0;
}
Exercise 7. Tilty Shift
This example was based on the addweighted.cpp algorithm and answers to the proposed exercises at agostinhobritojr.github.io.
This program adds three more functions to the provided example. Through these functions, the tiltyshift.cpp program makes the following adjustments:
- Adjust the height of the focus centre of the image
- Adjust the decay of the blurred area
- Adjust the vertical position of the focus centre area of the filtered image
#include <iostream>
#include <opencv2/opencv.hpp>
#include <cmath>
using namespace cv;
using namespace std;
double alfa;
int center_slider=0;
int center_slider_max=100;
int alfa_slider = 0;
int alfa_slider_max = 100;
// height
int top_slider = 0;
int top_slider_max = 100;
Mat image1, image2, blended;
Mat imageTop;
char TrackbarName[50];
void on_trackbar_blend(int, void*){
alfa = (double) alfa_slider/alfa_slider_max ;
addWeighted( image2, alfa, imageTop, 1-alfa, 0.0, blended);
imshow("addweighted", blended);
}
void on_trackbar_line(int, void*){
image2.copyTo(imageTop);
int width = image2.size().width;
int height = image2.size().height;
int limit = top_slider*height/100;
int base = center_slider*height/100;
if(limit > 0){
if(base >= 0 && base <= height-limit){
Mat tmp = image1(Rect(0, base, width,limit));
tmp.copyTo(imageTop(Rect(0, base, width,limit)));
}
else{
Mat tmp = image1(Rect(0, 0, width,limit));
tmp.copyTo(imageTop(Rect(0, 0, width,limit)));
}
}
on_trackbar_blend(alfa_slider,0);
}
int main(int argvc, char** argv){
image1 = imread("img/sydney7.jpg");
resize(image1,image1,Size(640,480));
image1.copyTo(image2);
namedWindow("addweighted", 1);
image2.convertTo(image2,CV_32F);
float media[]={1,1,1,
1,1,1,
1,1,1};
Mat mascara;
mascara = Mat(3,3,CV_32F,media);
scaleAdd(mascara, 1/9.0, Mat::zeros(3,3,CV_32F), mascara);
for (int i = 0; i < 7; i++) {
filter2D(image2, image2, image2.depth(), mascara, Point(1,1), 0);
}
image2.convertTo(image2, CV_8U);
image2.copyTo(imageTop);
sprintf( TrackbarName, "Decay x %d", alfa_slider_max );
createTrackbar( TrackbarName, "addweighted",
&alfa_slider,
alfa_slider_max,
on_trackbar_blend );
on_trackbar_blend(alfa_slider, 0 );
sprintf( TrackbarName, "Hight x %d", top_slider_max );
createTrackbar( TrackbarName, "addweighted",
&top_slider,
top_slider_max,
on_trackbar_line );
on_trackbar_line(top_slider, 0 );
sprintf( TrackbarName, "Center x %d", top_slider_max );
createTrackbar( TrackbarName, "addweighted",
¢er_slider,
center_slider_max,
on_trackbar_line );
on_trackbar_line(center_slider, 0 );
waitKey(0);
imwrite("tiltyshift.jpg", blended);
imshow("window",blended);
imwrite("original.jpg", image1);
imshow("window2",image1);
return 0;
}
Original image
Filtered image
Unit 2
In this unit, the following examples were developed:
- Homomorphic filtering
- Pointillism art
- The K-means process
Exercise 1. Homomorphic filter
This example implements an homomorphic filter based on the dft.cpp algorithm and answers to the proposed exercises at agostinhobritojr.github.io.
The Homomorphic filter is based on the following function:
What does this program do?
The program allows 4 different adjustments in the image
- Regulate the low frequency component γLγL (referring to illumination);
- Regulate the high frequency component γHγH (referring to reflectance);
- Regulate the variable C, which controls the slope of the function as it transitions between γLγL and γHγH.
- Adjust the variable D0D0;
Original image
Filtered image
#include <iostream>
#include <opencv2/opencv.hpp>
#include <cmath>
using namespace cv;
using namespace std;
Mat image, imageDft, padded;
char *filename;
float MAX = 100.0;
//-----------------VARIAVEIS DO FILTRO---------------------
float yl = 0, yh = 0, d0 = 0, c = 0;
float ylmax = 100, yhmax = 100, d0max = 256, cmax = 100;
int yl_slider = 0, yh_slider = 0, d0_slider = 0, c_slider = 0;
//---------------------------------------------------------
//valores ideais dos tamanhos da imagem para calculo da DFT
int dft_M, dft_N;
//---------------------------------------------------------
char TrackbarName[50];
void deslocaDFT(Mat& image ){
Mat tmp, A, B, C, D;
// se a imagem tiver tamanho impar, recorta a regiao para
// evitar cÃģpias de tamanho desigual
image = image(Rect(0, 0, image.cols & -2, image.rows & -2));
int cx = image.cols/2;
int cy = image.rows/2;
// reorganize quadrants
// A B -> D C
// C D B A
A = image(Rect(0, 0, cx, cy));
B = image(Rect(cx, 0, cx, cy));
C = image(Rect(0, cy, cx, cy));
D = image(Rect(cx, cy, cx, cy));
// A <-> D
A.copyTo(tmp); D.copyTo(A); tmp.copyTo(D);
// C <-> B
C.copyTo(tmp); B.copyTo(C); tmp.copyTo(B);
}
void filtroHM(){
Mat filter = Mat(padded.size(), CV_32FC2, Scalar(0));
Mat tmp = Mat(dft_M, dft_N, CV_32F);
for (int i = 0; i < dft_M; i++) {
for (int j = 0; j < dft_N; j++) {
float d2 = pow(i - dft_M/2.0, 2) + pow(j - dft_N/2.0, 2);
float exp = - c*(d2/pow(d0, 2));
float filtroH = (yh - yl)*(1 - expf(exp) ) + yl;
tmp.at<float> (i,j) = filtroH;
}
}
Mat comps[] = {tmp, tmp};
merge(comps, 2, filter);
Mat dftClone = imageDft.clone();
mulSpectrums(dftClone,filter,dftClone,0);
deslocaDFT(dftClone);
idft(dftClone, dftClone);
vector<Mat> planos;
split (dftClone, planos);
normalize(planos[0], planos[0], 0, 1, CV_MINMAX);
imshow("Filtro Homomorfico", planos[0]);
imshow("Original", image);
}
void on_trackbar_yl(int, void*){
yl = (float) yl_slider;
yl = ylmax*yl/MAX;
filtroHM();
}
void on_trackbar_d0(int, void*){
d0 = d0_slider*d0max/MAX;
filtroHM();
}
void on_trackbar_yh(int, void*) {
yh = yh_slider*yhmax/MAX;
filtroHM();
}
void on_trackbar_c(int, void*) {
c = c_slider*cmax / MAX;
filtroHM();
}
int main(int argvc , char** argv){
//open image
if (argvc != 2) {
cerr << "Usage: " << argv[0] << " <img_path>" << endl;
return 1;
}
filename = argv[1];
image = imread(filename);
cvtColor(image, image, CV_BGR2GRAY);
// Identify the optimum size for
// Calculating FFT
dft_M = getOptimalDFTSize(image.rows);
dft_N = getOptimalDFTSize(image.cols);
// Image padding
Mat_<float> zeros;
copyMakeBorder(image, padded, 0,
dft_M - image.rows, 0,
dft_N - image.cols,
BORDER_CONSTANT, Scalar::all(0));
// Zero padding - complex part of the matrix
zeros = Mat_<float>::zeros(padded.size());
// prepare the complex part
imageDft = Mat(padded.size(), CV_32FC2, Scalar(0));
copyMakeBorder(image, padded, 0,
dft_M - image.rows, 0,
dft_N - image.cols,
BORDER_CONSTANT, Scalar::all(0));
Mat_<float> realInput = Mat_<float>(padded);
// insere the two components within the matrices arrays
vector<Mat> planos;
planos.push_back(realInput);
planos.push_back(zeros);
// combine matrices array into a single
// complex component
merge(planos, imageDft);
// calcula o dft
dft(imageDft, imageDft);
deslocaDFT(imageDft);
namedWindow("Homomorfic filter", WINDOW_NORMAL);
sprintf( TrackbarName, "Yh");
createTrackbar( TrackbarName, "Filtro Homomorfico",
&yh_slider,
MAX,
on_trackbar_yh);
sprintf( TrackbarName, "YL");
createTrackbar( TrackbarName, "Filtro Homomorfico",
&yl_slider,
MAX,
on_trackbar_yl);
sprintf( TrackbarName, "D0");
createTrackbar( TrackbarName, "Filtro Homomorfico",
&d0_slider,
MAX,
on_trackbar_d0 );
sprintf(TrackbarName, "C");
createTrackbar( TrackbarName, "Filtro Homomorfico",
&c_slider,
MAX,
on_trackbar_c);
filtroHM();
waitKey(0);
return 0;
}
git
Exercise 2. Pointillism art
This section uses Canny Edge Detection, developed by John F. Canny in 1986, to create pointillism arts. To achieve this goal, we can simulate a painting by drawing circles that work out as a pointillism art. The example below perform the desired effect of pointillism through image processing.
The following steps were used to accomplish this effect:
What does this program do?
The program allows 4 different adjustments in the image
- Apply canny algorithm on an image.
- Use the edges identified by the canny algorithm to draw smaller circles on them;
- Uses the Canny threshold to increase the quality of the painting affect.
This example was based on the pointilhismo.cpp and canny.cpp algorithms and it answers to the proposed exercises at [agostinhobritojr.github.io]https://agostinhobritojr.github.io/tutorial/pdi/#_exerc%C3%ADcios_7).
Original image
Canny edge detector
Filtered image
#include <iostream>
#include "opencv2/opencv.hpp"
#include <fstream>
#include <iomanip>
#include <vector>
#include <algorithm>
#include <numeric>
#include <ctime>
#include <cstdlib>
using namespace std;
using namespace cv;
#define STEP 4
#define JITTER 2
#define RAIO 5
int top_slider = 10;
int top_slider_max = 200;
char TrackbarName[50];
Mat image,imgray, border,points;
int width, height;
Vec3b colors;
int x, y;
vector<int> yrange;
vector<int> xrange;
void on_trackbar_canny(int, void*){
Canny(imgray, border, top_slider, 3*top_slider);
imshow("cannyborders.png", border);
points = Mat(height, width, CV_8UC3, Scalar(255,255,255));
std::random_shuffle(xrange.begin(), xrange.end());
for(auto i : xrange){
random_shuffle(yrange.begin(), yrange.end());
for(auto j : yrange){
if(border.at<uchar>(j,i) == 255)
{
x = i+rand()%(2*JITTER)-JITTER+1;
y = j+rand()%(2*JITTER)-JITTER+1;
colors = image.at<Vec3b>(y,x);
circle(points, Point(x,y),2, CV_RGB(colors[2],colors[1],colors[0]), -1, CV_AA);
}
else{
x = i+rand()%(2*JITTER)-JITTER+1;
y = j+rand()%(2*JITTER)-JITTER+1;
colors = image.at<Vec3b>(x,y);
circle(points,
cv::Point(y,x),
RAIO,
CV_RGB(colors[2],colors[1],colors[0]),
-1,
CV_AA);
}
}
}
imshow("canny",points);
}
int main(int argc, char**argv){
image= imread(argv[1],CV_LOAD_IMAGE_COLOR);
cvtColor(image,imgray, CV_BGR2GRAY);
srand(time(0));
if(!image.data){
cout << "nao abriu" << argv[1] << endl;
cout << argv[0] << " imagem.jpg";
exit(0);
}
width=image.size().width;
height=image.size().height;
xrange.resize(height/STEP);
yrange.resize(width/STEP);
iota(xrange.begin(), xrange.end(), 0);
iota(yrange.begin(), yrange.end(), 0);
for(uint i=0; i<xrange.size(); i++){
xrange[i]= xrange[i]*STEP+STEP/2;
}
for(uint i=0; i<yrange.size(); i++){
yrange[i]= yrange[i]*STEP+STEP/2;
}
sprintf( TrackbarName, "Threshold_inferior", top_slider_max );
namedWindow("canny",1);
createTrackbar( TrackbarName, "canny",
&top_slider,
top_slider_max,
on_trackbar_canny );
on_trackbar_canny(top_slider, 0 );
waitKey();
imwrite("cannypontos.png",points);
return 0;
}
Exercise 3. K-means proccess
This example was based on the kmeans.cpp algorithm and answers to the proposed exercises at [agostinhobritojr.github.io]https://agostinhobritojr.github.io/tutorial/pdi/#_exerc%C3%ADcios_8).
The following steps were used to accomplish this effect:
What does this program do?
The k-means algorithm works according to the following steps:
- Choose k as the number of classes for the vectors xi of N samples, i = 1,2, ⋯, N.
- Choose m1, m2, ⋯, mk as initial approximations for the class centers.
- Sort each sample xi using, for example, a minimum distance classifier (Euclidean distance).
- Recalculate the averages mj using the result of the previous step.
- If the new averages are consistent (do not change considerably), finalize the algorithm. If not, recalculate the centers and redo the classification.
#include <opencv2/opencv.hpp>
#include <cstdlib>
#include <ctime>
using namespace cv;
int main( int argc, char** argv ){
for(int i = 0; i < 10; i++){
int nClusters = 20;
Mat labels;
int nRounds = 1;
Mat centers;
char name[30];
if(argc!=3){
exit(0);
}
Mat img = imread( argv[1], CV_LOAD_IMAGE_COLOR);
Mat samples(img.rows * img.cols, 3, CV_32F);
for( int y = 0; y < img.rows; y++ ){
for( int x = 0; x < img.cols; x++ ){
for( int z = 0; z < 3; z++){
samples.at<float>(y + x*img.rows, z) = img.at<Vec3b>(y,x)[z];
}
}
}
kmeans(samples,
nClusters,
labels,
TermCriteria(CV_TERMCRIT_ITER|CV_TERMCRIT_EPS, 10000, 0.0001),
nRounds,
KMEANS_RANDOM_CENTERS,
centers );
Mat labeled( img.size(), img.type() );
for( int y = 0; y < img.rows; y++ ){
for( int x = 0; x < img.cols; x++ ){
int indice = labels.at<int>(y + x*img.rows,0);
labeled.at<Vec3b>(y,x)[0] = (uchar) centers.at<float>(indice, 0);
labeled.at<Vec3b>(y,x)[1] = (uchar) centers.at<float>(indice, 1);
labeled.at<Vec3b>(y,x)[2] = (uchar) centers.at<float>(indice, 2);
}
}
sprintf(name, "test%d.jpg", i);
imwrite(name, labeled);
//waitKey( 0 );
}
}