Image clustering using Python OpenCV

git clone https://github.com/xsyann/clustering.git

python clustering.py

cv2, numpy, matplotlib, scipy, urllib2, PyQt5

Matplotlib Qt5 compatibility: matplotlib/matplotlib#3072

Elbow

• Draw the green line l passing through the first and the last point of the blue curve.
• Select the maximum length of the red segment perpendicular to the line l and passing through the curve point.

Threshold

• Reduce the compactness in range [0-1].
• Select the first point below the threshold defined in the Clusterer class.

Gap statistic

• Gap statistic by Robert Tibshirani, Guenther Walther and Trevor Hastie.
• Stanford University, USA, November 2000.

Gap Threshold

• Reduce the compactness in range [0-1].
• Compute the gap between each neighbors point.
• Select the first point where the gap is below the threshold defined in the Clusterer class.

f(k)

• Selection of K in K-means clustering, D T Pham�, S S Dimov, and C D Nguyen.
• Manufacturing Engineering Centre, Cardiff University, Cardiff, UK, 27 September 2004.

Slider

• Reduce the compactness and the cluster count in range [0-1].
• Draw a line between (0, 1) and (1, 0) and place a goal on this line depending of a percentage.
• Select the closest point to the goal.

One clustering feature

Two clustering features

Three clustering features

• http://blog.echen.me/2011/03/19/counting-clusters/
• https://gist.github.com/michiexile/5635273
• http://www.stanford.edu/~hastie/Papers/gap.pdf
• http://datasciencelab.wordpress.com/2014/01/21/selection-of-k-in-k-means-clustering-reloaded/
• http://www.ee.columbia.edu/~dpwe/papers/PhamDN05-kmeans.pdf
• http://pami.uwaterloo.ca/groups/ras/PAMI_presentation.pdf
• http://www.ima.umn.edu/~iwen/REU/paper4.pdf
• http://docs.opencv.org/modules/core/doc/clustering.html
• http://stackoverflow.com/questions/10650645/python-calculate-voronoi-tesselation-from-scipys-delaunay-triangulation-in-3d