A Corner Detector based on Global and Local Curvature Properties

A robust corner detector works on grey-level images as well as planar curves.

Description

CORNER works by the following step:
1. Apply the Canny edge detector to the grey level image and obtain a binary edge-map.
2. Extract the edge contours from the edge-map, fill the gaps in the contours.
3. Compute curvature at a low scale for each contour to retain all true corners.All of the curvature local maxima are considered as initial corner candidates.
4. Initial corner candidates are compared using an adaptive local threshold to remove the round corners.
5. The angles of corner candidates are evaluated to eliminate any false corners due to quantization noise and trivial details. The above evaluation is based on a dynamic region of support, which changes according to its adjacent corner candidates.
6. End points of contours are considered with an additional criterion.

Algorithm is derived from:
X.C. He and N.H.C. Yung, "Curvature Scale Space Corner Detector with Adaptive Threshold and Dynamic Region of Support", Proceedings of the 17th International Conference on Pattern Recognition, 2:791-794, August 2004.

Improved algorithm of above has been published in:
X.C. He and N.H.C. Yung, “Corner detector based on global and local curvature properties”, Optical Engineering, 47(5), pp: 057008, 2008. (This paper is included in corner.rar for reference)

• Acknowledgements: This submission has inspired the following:; Robust Image Corner Detection based on the Chord-to-Point Distance Accumulation Technique
• Required Products: Image Processing Toolbox
• MATLAB release: MATLAB 6.5 (R13)
• Other Files: Corner.rar, ; license.txt

[Itachi Uchiha]

Great work! I've tried many corner detectors (Harris-Plessey, SUSAN, Kitchen/Rosenfeld...) and this one gives the best results. Highly recommended!

[Youn Kim]

I like the "extract_curve" function in this code.

[Shripad Kondra]

Simply gr8 ! This is a nice algorithm, easy to understand, specially for newbies in contour tracing

[Zhigang Tan]

it integrates both local and global information to determine a salient corner. Just behave as a multi-scale corner detector. Really great work! Though better mathematical explanation would reputate it.

[Mohammad Awrangjeb]

simply gr8!

[abu bakar elmi]

I had tried and it work great.

[Don M]

Excellent job. Code works as perfect. Thank you for your great work!

[ming cui]

Well ,the result is pretty good, but the author did not do the multiscale tracing, claimed as step 5 & 6 in his icpr paper. If this programmme is used to prove the paper, then it's kind of incomplete.

[gorkem serbes]

It works perfect, nice job.

[Xiaochen He]

Reply to ming cui's comments: the multiscale tracing is not a step of my corner detector. It is adopted by CSS corner detector, which was referenced in my ICPR paper.

[Daniel Daniel]

 

[Nan Jiang]

The flip on retrieving zero-crossing point is smart.
In Fit a circle step, what (x0, y0) suppose to mean? I guess it the center of the fitted circle? Am I right?
Thanks.

[rainer martin]

good work and thansk for this publication

[Jonathan Perilla]

 

[Nat Ang]

Good code, works fine. Just a little addition...if the image (like in some of my cases) does not have corners the program produces an error...solved by adding in the function get_corner another conditional in the last if...
add
if count,
  if Endpoint,....

Other than that it's perfect. The error was generated when evaluating segments of ellipsoids and looking for corners in the edges to identify overlapping ellipses.
If the segment did not have corners neither was closes (a full ellipse) then the error was generated....this is for general information it might be useful for someone!.

THIS CODE SAVE ME A LOT OF TIME!!!
THANKS

[Somaya Mostafa]

Good code for making edge and corner detection

[z lijie]

thank you very much!
Excellent!

[minde min daugelis]

 

[Suvdaa B]

Thank you very much.
Good luck for your research work.

[liu YY]

 

[hbu ]

good ,thank you very much!

[Jozs]