Obtaining list of unique pixel values in OpenCV Mat - c++

Question

Is there an equivalent of np.unique() or bincount() for an OpenCV Mat? I am working with C++ so can't just convert to a numpy array.

Answer 1

No, there is not! You can code your own, though:
std::vector<float> unique(const cv::Mat& input, bool sort = false)
Find the unique elements of a single channel cv::Mat.
Parameters:
input: It will be treated as if it was 1-D.
sort: Sorts the unique values (optional).
The implementation of such function is pretty straight forward, however, the following only works with single channel CV_32F:
#include <algorithm>
#include <vector>
std::vector<float> unique(const cv::Mat& input, bool sort = false)
{
if (input.channels() > 1 || input.type() != CV_32F)
{
std::cerr << "unique !!! Only works with CV_32F 1-channel Mat" << std::endl;
return std::vector<float>();
}
std::vector<float> out;
for (int y = 0; y < input.rows; ++y)
{
const float* row_ptr = input.ptr<float>(y);
for (int x = 0; x < input.cols; ++x)
{
float value = row_ptr[x];
if ( std::find(out.begin(), out.end(), value) == out.end() )
out.push_back(value);
}
}
if (sort)
std::sort(out.begin(), out.end());
return out;
}
Example:
float data[][3] = {
{ 9.0, 3.0, 7.0 },
{ 3.0, 9.0, 3.0 },
{ 1.0, 3.0, 5.0 },
{ 90.0, 30.0, 70.0 },
{ 30.0, 90.0, 50.0 }
};
cv::Mat mat(3, 5, CV_32F, &data);
std::vector<float> unik = unique(mat, true);
for (unsigned int i = 0; i < unik.size(); i++)
std::cout << unik[i] << " ";
std::cout << std::endl;
Outputs:
1 3 5 7 9 30 50 70 90

Answer 2

You could try to build a histogram with number of bins equal to number of possible pixel values.

Answer 3

Here is another suggestion for an implementation using the standard library.
opencv-unique.cpp
#include <opencv2/opencv.hpp>
#include <vector>
#include <iostream>
#include <algorithm>
#include <cstdint>
/**
* #brief Find unique elements of an OpenCV image
*
* #tparam type is the C++ type to access the image elements.
* #param in is the OpenCV single-channel image to find the unique values. Note: This
* modifies the image. Make a copy with .clone(), if you need the image afterwards.
*
* #returns vector of unique elements
*/
template<typename type>
std::vector<type> unique(cv::Mat in) {
assert(in.channels() == 1 && "This implementation is only for single-channel images");
auto begin = in.begin<type>(), end = in.end<type>();
auto last = std::unique(begin, end); // remove adjacent duplicates to reduce size
std::sort(begin, last); // sort remaining elements
last = std::unique(begin, last); // remove duplicates
return std::vector<type>(begin, last);
}
int main() {
cv::Mat img = (cv::Mat_<uint16_t>(3, 4) << 1, 5, 3, 4, 3, 1, 5, 5, 1, 3, 4, 3);
std::cout << "unique values: ";
auto u = unique<uint16_t>(img);
for (auto v : u)
std::cout << v << " ";
std::cout << std::endl;
return 0;
}
Compile and execute yields:
$ g++ -Wall opencv-unique.cpp -o unique -lopencv_core -I /usr/include/opencv4 && ./unique
unique values: 1 3 4 5
The version above is for single-channel images. You can extend this to multi-channel images (to get unique colors), like this.