extract a mat which have address in a mat from another mat

edit

I have done a little example, so there are some things to notice:

(i.) In Matlab the index starts with 1, in C++ it starts with 0

(ii.) 'b' is your array of indices for an 1D array

(iii.) In OpenCV the read out of the pixels is row by row, in Matlab it seems to be column by column

// include header for OpenCV3
#include <opencv2/opencv.hpp>

using namespace cv;

int main() {
    // Fill arrays by hand, this is just for this example
    Mat A(3, 3, CV_32SC1);
    A.at<int>(0, 0) = 4;
    A.at<int>(1, 0) = 10;
    A.at<int>(2, 0) = 5;
    A.at<int>(0, 1) = 9;
    A.at<int>(1, 1) = 3;
    A.at<int>(2, 1) = 8;
    A.at<int>(0, 2) = 8;
    A.at<int>(1, 2) = 7;
    A.at<int>(2, 2) = 4;

    Mat b(2, 2, CV_32SC1);
    // Note that the indices are decreased by 1 (see above)
    b.at<int>(0, 0) = 1;
    b.at<int>(1, 0) = 2;
    b.at<int>(0, 1) = 3;
    b.at<int>(1, 1) = 4;

    // Transpose array, because of row by row problem
    A = A.t();
    b = b.t();

    // Reshape arrays from 2D to 1D. Note that the 1 is not for dimension, but for number of rows (9 cols, 1 row)
    A = A.reshape(0, 1);
    b = b.reshape(0, 1);

    // Creating 1D Result array
    Mat ans(b.size(), b.type());

    // Fill the result array with A at the indices from b
    for(int i=0; i<b.cols; i++) {
        ans.at<int>(0, i) = A.at<int>( 0, b.at<int>(0, i) );
    }

    // Reshape it back to 2D. The 2 is for the number of rows: 2 cols, 2 rows
    ans = ans.reshape(0, 2);
    // Transpose it back, because of OpenCV conventions
    ans = ans.t();

    // Print out the result
    for(int i=0; i<ans.rows; i++) {
        for(int j=0; j<ans.cols; j++) {
            cout << (int)ans.at<int>(j, i) << " ";
        }
        cout << endl;
    }
return 0;
}

For matrix operations OpenCV is not the first choice. Take a look at the Eigen libraries. In OpenCV there is also a module to share data with Eigen.

EDIT: Answer to the comment below: I recommend you to use cv::InputArray instead of cv::Mat to pass arrays in functions. Check out InputArray.

The sample code is just an example (I hope I did no mistake, I could not check this at the moment) and you have to catch some more cases if needed, for example that the index value doesn't exceed the array limits,... Take a look in the source code from OpenCV to learn more about InputArrays and how to proof it.

// Use cv::InputArray, cv::OutputArray or cv::InputOutputArray
cv::Mat extract(cv::InputArray SRC, cv::InputArray IDX, bool isIdxContinous)
{
    // check that IDX format is of type std::vector or cv:Mat
    cv::CV_Assert(IDX.kind() _InputArray::STD_VECTOR || IDX.kind() == _InputArray::MAT);

    // check that IDX is of type int (You should convert ...