Implement Harris score for FAST corners

[opatiny]

Harris is actually a corner detector, so how do we make to only get the Harris score? https://en.wikipedia.org/wiki/Harris_corner_detector#:~:text=The%20Harris%20corner%20detector%20is,improvement%20of%20Moravec's%20corner%20detector.

Harris original article: http://www.bmva.org/bmvc/1988/avc-88-023.pdf

[opatiny]

• I have tried implementing the Harris score, but I really struggle with testing the results to verify that what I get is not completely irrelevant. Also, it feels weird to me to use Harris only as a score of a currently detected corner, because the original algorithm is actually used as a corner detector itself.
• Second issue: the algorithm is really slow when used as an option to compute the scores in getFastKeypoints. This should be fixed.

[opatiny]

Update on my last message:

• I have written a Matlab script to generate FAST features of an image in order to compare with my implementation. The issue is that the option for Harris score does not exist in Matlab...
• Performance issues: They were mainly caused by me saving little images for debug, which really slowed down the code. I removed the lines saving the images and it's already a lot better (similar performance to cannyEdgeDetector()

[opatiny]

Matlab script and results

FAST

%% Detect FAST corners in an image
clc; clear; close all;

%% load image
image = imread('alphabet.jpg');
image = rgb2gray(image);

%% detect corners
corners = detectFastFeatures(image);

%% display results
imshow(image); hold on;
p = plot(corners.selectStrongest(50));
hold off;

Harris corner detector (not score)

%% Detect Harris corners in an image
clc; clear; close all;

%% load image
image = imread('alphabet.jpg');
image = rgb2gray(image);

%% detect corners
corners = detectHarrisFeatures(image);

%% display results
imshow(image); hold on;
 p = plot(corners.selectStrongest(50));
hold off;