How to thresholding image by ourselves without using fucntion in OpenCV - opencv

Question

I'm new in OpenCV, and I want to thresholding the image by myself without using Threshold function in opencv, because the time spend on function threshold is to high for me.
Here is my code:
Mat src = imread("D:\\DataBox\\7.jpg", 0);
for (int i = 0; i < src.cols; i++) {
cout << i << endl;
for (int j = 0; j < src.rows; j++) {
if (src.at<uchar>(i, j) > 70) {
src.at<uchar>(i, j) = 0;
cout << j << endl;
}
else
src.at<uchar>(i, j) = 255;
}
}
but it still says:
"OpenCV Error: Assertion failed (dims <= 2 && data && (unsigned)i0 < (unsigned)size.p[0] && (unsigned)(i1 * DataType<_Tp>::channels) < (unsigned)(size.p[1] * channels()) && ((((sizeof(size_t)<<28)|0x8442211) >> ((DataType<_Tp>::depth) & ((1 << 3) - 1))*4) & 15) == elemSize1()) in cv::Mat::at, file C:\Program Files\opencv\build\include\opencv2/core/mat.inl.hpp, line 894"
I can print j from 0~719(since the size of the image is 720*960), but as long as the parameter i want to become 2 from 1, the error occurs.

Answer 1

You mixed up rows and cols:
Try this:
Mat src = imread("path_to_image", IMREAD_GRAYSCALE);
for (int i = 0; i < src.rows; i++)
{
//cout << i << endl;
for (int j = 0; j < src.cols; j++)
{
if (src.at<uchar>(i, j) > 70) {
src.at<uchar>(i, j) = 0;
//cout << j << endl;
}
else
src.at<uchar>(i, j) = 255;
}
}
This is, however very unlikely to perform better than OpenCV implementation. You can gain a little speed working on raw pointers, with a little trick to work on continuous data when possible:
#include <opencv2\opencv.hpp>
using namespace cv;
int main()
{
Mat src = imread("D:\\SO\\img\\nice.jpg", IMREAD_GRAYSCALE);
int rows = src.rows;
int cols = src.cols;
if (src.isContinuous())
{
cols = rows * cols;
rows = 1;
}
for (int i = 0; i < rows; i++)
{
uchar* pdata = src.ptr<uchar>(i);
int base = i*cols;
for (int j = 0; j < cols; j++)
{
if (pdata[base + j] > 70)
{
pdata[base + j] = 0;
}
else
{
pdata[base + j] = 255;
}
}
}
return 0;
}
Actually, on my PC my version is a little bit faster than OpenCV one:
Time #HenryChen (ms): 2.83266
Time #Miki (ms): 1.09597
Time #OpenCV (ms): 2.10727
You can test on your PC with the following code, since time depends on many factor, e.g. optimizations enabled in OpenCV:
#include <opencv2\opencv.hpp>
#include <iostream>
using namespace cv;
using namespace std;
int main()
{
Mat1b src(720,960);
randu(src, 0, 256);
Mat1b src1 = src.clone();
Mat1b src2 = src.clone();
Mat1b src3 = src.clone();
double tic1 = double(getTickCount());
// Method #HenryChen (corrected)
for (int i = 0; i < src1.rows; i++)
{
//cout << i << endl;
for (int j = 0; j < src1.cols; j++)
{
if (src1.at<uchar>(i, j) > 70) {
src1.at<uchar>(i, j) = 0;
//cout << j << endl;
}
else
src1.at<uchar>(i, j) = 255;
}
}
double toc1 = (double(getTickCount()) - tic1) * 1000.0 / getTickFrequency();
cout << "Time #HenryChen (ms): \t" << toc1 << endl;
//-------------------------------------
double tic2 = double(getTickCount());
// Method #Miki
int rows = src2.rows;
int cols = src2.cols;
if (src2.isContinuous())
{
cols = rows * cols;
rows = 1;
}
for (int i = 0; i < rows; i++)
{
uchar* pdata = src2.ptr<uchar>(0);
int base = i*cols;
for (int j = 0; j < cols; j++)
{
pdata[base + j] = (pdata[base + j] > 70) ? uchar(0) : uchar(255);
}
}
double toc2 = (double(getTickCount()) - tic2) * 1000.0 / getTickFrequency();
cout << "Time #Miki (ms): \t" << toc2 << endl;
//-------------------------------------
double tic3 = double(getTickCount());
// Method #OpenCV
threshold(src3, src3, 70, 255, THRESH_BINARY_INV);
double toc3 = (double(getTickCount()) - tic3) * 1000.0 / getTickFrequency();
cout << "Time #OpenCV (ms): \t" << toc3 << endl;
getchar();
return 0;
}

Answer 2

Use test.at<uchar>(cv::Point(i, j)) instead. I always get lost when accessing cv::Mat directly - cv::Point clears it up a little bit.
Anyway, I agree with Miki - it is very unlikely to create a function that performs better that a library one.