Refines the corner locations

corners = cv.cornerSubPix(im, corners)
corners = cv.cornerSubPix(im, corners, 'OptionName', optionValue, ...)

Input

• im Input single-channel image.
• corners Initial coordinates of the input corners, stored in numeric array (Nx2/Nx1x2/1xNx2) or cell array of 2-element vectors ({[x,y], ...}). Supports single floating-point class.

Output

• corners Output refined coordinates, of the same size and type as the input corners (numeric or cell matching the input format).

Options

• WinSize Half of the side length of the search window. For example, if WinSize=[5,5], then a (5 * 2 + 1) x (5 * 2 + 1) = 11 x 11 search window is used. default [3, 3].
• ZeroZone Half of the size of the dead region in the middle of the search zone over which the summation in the formula below is not done. It is used sometimes to avoid possible singularities of the autocorrelation matrix. The value of [-1,-1] indicates that there is no such a size. default [-1,-1].
• Criteria Criteria for termination of the iterative process of corner refinement. That is, the process of corner position refinement stops either after criteria.maxCount iterations or when the corner position moves by less than criteria.epsilon on some iteration. Default to struct('type','Count+EPS', 'maxCount',50, 'epsilon',0.001). Struct with the following fields is accepted:
  • type one of 'Count', 'EPS', or 'Count+EPS' to indicate which criteria to use.
  • maxCount maximum number of iterations
  • epsilon minimum difference in corner position

The function iterates to find the sub-pixel accurate location of corners or radial saddle points.

Sub-pixel accurate corner locator is based on the observation that every vector from the center q to a point p located within a neighborhood of q is orthogonal to the image gradient at p subject to image and measurement noise. Consider the expression:

epsilon_i = DI_p_i' * (q - p_i)

where DI_p_i is an image gradient at one of the points p_i in a neighborhood of q. The value of q is to be found so that epsilon_i is minimized. A system of equations may be set up with epsilon_i set to zero:

Sigma_i(DI_p_i * DI_p_i') - Sigma_i(DI_p_i * DI_p_i' * p_i)

where the gradients are summed within a neighborhood ("search window") of q. Calling the first gradient term G and the second gradient term b gives:

q = inv(G) * b

The algorithm sets the center of the neighborhood window at this new center q and then iterates until the center stays within a set threshold.