top of page
Lab. LGI2P
وزار Ø© التعليم العالي والبØØ« العلمي - العراق
الجامعة التقنية الشمالية
الدكتور Øسن كريم عبد الرØمن
اختصاص معالجة الصور الرقمية وامنية المعلومات
Corner and Junction Detection Method
In digital images, corners and junctions get affluent information like delimitation of objects, they constitute important landmarks.
Many algorithms for detecting the corners have been developed up to now. Indeed, corner detection is often an important step in various image-understanding and scene-reconstructing systems.
The corner detectors should satisfy the following criteria:
-
All the true corners and no false corners should be detected.
​
-
The corners should be located precisely.
​
-
The detectors should not be sensitive to noise in images. very point.
PROPOSED METHOD EXPERIMENTS
we proposed new robust corner detection method and our experiments carried out on synthetic images and real image to compare corner detectors. Concerning the synthetic images, the signal is corrupted by a Gaussian noise over different Signal-to-Noise Ratios (SNR). Then, the Root-Mean-Square Error (RMSE) is computed between the 31 true corners and the 31 ”best” corners extracted by the corner detectors. Our proposed approach remains more stable and robust to noise than the nine compared methods.
Our method was implemented and tested. The following was found out:
-
The detection of corners is highly reliable.
​
-
The localization errors are small.
​
-
The sensitivity to noise is low.
PROPOSED METHOD RESULTS WITH = 1
Experiments results with synthetic image
- Our proposed method remains more stable and robust to noise than the nine compared methods. I published the two close results to our method results.
PROPOSED METHOD
Aach method [1]
True corner = + , False corner = +, Forget corner = +.
Shui method [2]
SNR= , RSME=0.000
SNR= , RSME=0.4399
SNR= , RSME=0.9672
SNR=11 dB, RSME=0.3113
SNR=10 dB, RSME=0.3592
SNR= 9 dB, RSME=0.3111
SNR=11 dB, RSME=0.4751
SNR=10 dB, RSME=0.5388
SNR= 9 dB, RSME=0.3119
SNR=11 dB, RSME=0.9672
SNR=10 dB, RSME=0.9673
SNR= 9 dB, RSME=0.9158
SNR= 8 dB, RSME=0.4399
SNR= 7 dB, RSME=0.4016
SNR= 6 dB, RSME=0.3593
SNR= 5 dB, RSME=0.5957
SNR= 4 dB, RSME=1.8050
SNR= 8 dB, RSME=0.4016
SNR= 7 dB, RSME=0.6476
SNR= 6 dB, RSME=0.5388
SNR= 5 dB, RSME=4.0521
SNR= 8 dB, RSME=1.0000
SNR= 7 dB, RSME=1.9610
SNR= 6 dB, RSME=3.3118
SNR= 5 dB, RSME=5.7712
SNR= 4 dB, RSME=5.8101
SNR= 4 dB, RSME=8.5647
Experiments results with real image
To make the comparison easier, only 180 best interest points are extracted for each method
PROPOSED METHOD
Kitchen method[3]
Fast method [6]
Aach method [1]
Shi-Tomasi method[4]
Forstner method [7]
Shui method [2]
Xia method[5]
Harris method [8]
Koethe method [9]
References
[1] T. Aach, C. Mota, I. Stuke, M. Muhlich, and E. Barth,“Analysis of superim oriented patterns,” IEEETIP, vol. 15, no. 12, pp. 3690–3700, 2006.
[2] P.-L. Shui and W.-C. Zhang, “Corner detection and classification using anisotropic directional derivative representations,” IEEE TIP, vol. 22, no. 8, pp. 3204–3218, 2013.
[3] L. Kitchen and A. Rosenfeld, “Gray-level corner detection,”
Tech. Rep., DTIC Document, 1980.
[4] J. Shi and C. Tomasi, “Good features to track,” in IEEE CVPR, 1994, pp. 593–600.
[5] G.S. Xia, J. Delon, and Y. Gousseau, “Accurate junction detection and characterization in natural images,” IJCV, vol. 106, no. 1, pp. 31–56, Springer, 2014.
[6]E. Rosten, R. Porter, and T. Drummond, “Faster and better: A machine learning approach to corner detection,”IEEE TPAMI, vol. 32, no. 1, pp. 105–119, 2010.
[7] W. Forstner, “A feature based correspondence algorithm for image matching,” International Archives of Photogrammetry and Remote Sensing, vol. 26, no. 3, pp. 150–166, 1986.
[8] C. Harris and M. Stephens, “A combined corner and edge detector.,” in Alvey Vision Conference, Citeseer, vol. 15, p. 50, 1988.
[9] U. K¨othe, “Edge and junction detection with an improved structure tensor,” in Joint Patt. Rec. Symp.,
Springer, pp. 25–32, 2003.
​
​
​
​
​
​
​
​
​
​
​
There are more experiments I will publish in the website later.
​
bottom of page