Halcon3: clustering and achieve halcon
(1) clustering profile:
Clustering is an unsupervised learning method, trying to sample data set is divided into several sub-sets usually do not intersect.
Two basic questions (2) clustering:
① performance metrics:
Quality metrics for clustering results, i.e. the same sample cluster as similar as possible, as different samples of different clusters. Performance of the star can be classified into two groups, the external indicators, indicators of the clustering result with an external comparison, two internal indicators, a direct inspection clustering results.
② distance calculation:
We used some form based on the concept of "distance" to define "similarity measure", the greater the distance, the smaller the similarity, but this "distance" may not satisfy all the essential nature of distance measures.
(3) prototype clusters:
① prototype clustering profile:
Prototype cluster configuration assumed by a set of cluster prototypes characterizations represent different prototype will produce different algorithms.
② several famous prototype clustering:
1) k-means clustering (k-means) [This is the way I want to cluster]
<1> k-means clustering Introduction:
K-means clustering average error minimization resulting clustering in the clustering, the average error is smaller, that is, the higher the similarity within a cluster sample.
<2> k-means clustering algorithm process is as follows:
Mean vector of input samples to initialize;
Sample concentration k samples randomly selected as an initial vector;
The clusters divided;
The mean vector iterative update, know the iterative updating mean vector unchanged, the current cluster division result output.
2) learning vector quantization (LVQ)
3) mixed Gaussian Clustering
(4) k-means clustering halcon achieved:
cluster model components: the new parameters used to create the model of a component for training results.
inspect clustered components: a rigid member model obtained from the training check.
class 2dim unsup: clustering segmentation in the two images.
class 2dim sup: two-dimensional image space pixel classification division.
histo 2dim: calculating two-channel histogram of grayscale images.