Type casting on the data pointer of openCV

Question

codes 1

template<typename T>
inline T* get_pointer(cv::Mat &src, int row)
{
    return src.ptr<T>(row);
}

template<typename T>
inline T* get_pointer(cv::Mat &src, int row, int col)
{
    return get_pointer<T>(src, row) + col * src.channels();
}

template<typename T>
inline T* get_pointer(cv::Mat &src, int row, int col, int channel)
{
    return get_pointer<T>(src, row, col) + channel;
}

codes 2

cv::Mat input = //....
auto *input_ptr = get_pointer<float>(input, row, col);
//back to previous row
input_ptr = reinterpret_cast<float*>(reinterpret_cast<uchar*>(input_ptr) - input.steps);

Are they safe?

1) reinterpret_cast<> is never "safe." (2) The pointer math in get_pointer(src, row) will fail for non-byte pixel types. (3) Are you aware of cv::Mat.ptr() methods? I think they do what you're trying to do — Mike C
– Mike C, Commented Mar 29, 2013 at 21:45
About the non byte pixel types, do you mean some channels like 5-6-5(3 channels with 16bits, looks like pretty command on embedded device).I always afraid of type casting on c++, especially the reinterpret cast, but I don't know other solution can go back to the previous row if I want to do it by pointer arithmetic.Thanks for your information of .ptr<T>(), I change the implementation already. — StereoMatching
– StereoMatching, Commented Mar 30, 2013 at 2:11
5-6-5, or 16-bit unsigned, or float. The pointer math needs to apply the reinterpret_cast<> to the data before being used in a calculation. I think you should not implement your own templates and simply use the cv::Mat methods directly. Not sure how OpenCV handles 5-6-5, tho. — Mike C
– Mike C, Commented Apr 1, 2013 at 18:08

Sameer · Accepted Answer · 2015-03-23 21:34:59Z

1

Why don't you use this (shorter and safer) code instead?

T *ptr_to_elem = &src.at<T>(row,col)[channel];

This works for non-contiguous arrays too. No potentially dangerous reinterpret_cast<> needed. The only condition for safety here is that you know the datatype T. A more safe version is when you use cv::Mat_<T> instead of cv::Mat, so that mistakes are caught at compile time instead of crashing your program at run time.

T *ptr_to_elem = &src(row,col)[channel];

answered Mar 23, 2015 at 21:34

Sameer

2,56523 silver badges13 bronze badges

Sign up to request clarification or add additional context in comments.

Collectives™ on Stack Overflow

Type casting on the data pointer of openCV

1 Answer 1

Comments

Your Answer

Hot Network Questions

Collectives™ on Stack Overflow

1 Answer 1

Comments

Your Answer

Sign up or log in

Post as a guest

Related