3.K-means clustering - kNN & K-means algorithm

kNN algorithm

1. What is kNN - Nearest Neighbor algorithm

• It is one of supervise train algorithm for classification or regression that look for train data of K number, where is mostly near by test data in feature space.

• NN vs kNN
  • NN : Nearest neighbor (k=1)

• KNN algorithm object creation

cv2.ml.KNearest_create() -> retval

• retval : cv2.ml_KNearest object

• Input data of classification using the KNN algorithm

cv.ml_KNearest.findNearest(samples, k, results=None, neighborResponses=None, dist=None, flgs=None) -> retval, results, neighborResposnes, dist

• samples : Input sample matric with input vector stored by row unit

• KNN algorithm poin classification example

2. KNN digit recognition

• If it is the printed digit to setted font, Template matching is possible

• KNN digit recognition processing
  • Making coordination of one point from 400-dimension using pixel value of image 20 x 20
  • KNN algorithm point classification in 400-dimension space

• KNN digit recognition flow chart

2. k-means algorithm

• k-means algorithm
  • Cluster algorithm the given data divides section of k number
• Process
  1. Select random k number center
  2. Select nearest center to every data
  3. Recalculate center to each cluster
  4. Repeat 2~3 process, If center changed
  5. If not, it’ll end

• Color image segment
  • Express each pixel value of input image to the one point of color space
  • Perform k-means algorithm in color space
  • Translate each pixel value into representive color of k number

• K-mean cluster code

cv2.kmeans(data, K, bestLabels, criteria, attempts, flags, centers=None) -> retval, bestLabels, centers

• data : train data matrix
• K : cluster number
• bestLabels : cluster set matrix of each sample
• criteria : finish standard
• attempts : repeating number to using another initial label
• flags : initial cent setting method
• centers : matrix for expressing the matrix
• retval : compatness measure

• k-means algorithm example