Learning OpenCV, 2nd Edition

M.at<float>( i, j );

Element ( i, j ) (row, col) from float

array M

M.at<int>( pt );

Element at location (pt.x, pt.y) in

integer matrix M

M.at<float>( i, j, k );

Element at location ( i, j, k ) in

three-dimensional float array M

M.at<uchar>( idx );

Element at n-dimensional location

indicated by idx[] in array M of

unsigned characters

To access a two-dimensional array, you can also extract a C-style pointer to a specific row of the array.

This is done with the ptr<>() template member function of cv::Mat. (Recall that the data in the array

is contiguous by row, thus accessing a specific column in this way would not make sense; we will see the

right way to do that shortly.) As with at<>(), ptr<>() is a template function instantiated with a type

name. It takes an integer argument indicating the row you wish a pointer to. The function returns a pointer

to the primitive type of which the array is constructed (i.e., if the array type is F32C3, the return value will

be of type float*). Thus, given a three-channel matrix mtx of type float, the construction

mtx.ptr<Vec3f>(3) would return a pointer to the first (floating-point) channel of the first element in

row 3 of mtx. This is generally the fastest way to access elements of an array,

18

because once you have the

pointer, you are right down there with the data.

There are thus two ways to get a pointer to the