Xiaolu Lixl76

Example of a right floating element.

Interest point detection

The Harris corner detector is implemented to find the basic local features.

scale	0.01	0.015	0.017	0.0175	0.018	0.02	0.03	0.04	0.05
wrong match	16	15	11	11	11	11	11	13	14
correct match	86	83	82	82	82	79	76	74	70
rate	0.186	0.181	0.134	0.134	0.134	0.139	0.145	0.17	0.20

Set scale = 0.018. Similarily, set rate of distance to 0.77.

Local feature description

The default feature_width is 16. We need to calculate the gradient of the 16 * 16 matrix surrounding the feature point.

for p = 1:4
    for q = 1:4
 	dy = aY(p,q);
        dx = aX(p,q);
        at = (pi + atan2(dy,dx))*4/pi;
        index = ceil(at);
        count(index) = count(index) + norm([dy dx]);
    end
end

Feature matching

Find the two nearest neighbors to the target feature, and then store the ratio as confidences. The matlab built-in function sort is very efficient in finding the nearest neighbors. The following is how I implemented.

for j = 1 :num_features2
        v = [v sqrt(sum((features1(i, :) - features2(j , :)).^2))];
    end
    [v ind]  = sort(v);
    d1 = v(1);
    d2 = v(2);
    minindex = ind(1);
    if d1 / d2 < 0.77
        matches = [matches ; [i minindex]];
        confidences = [confidences ; d1 / d2];
    end

Results