Different results for opencv c++ kmeans and matlab kmeans

Question

MATLAB K-means gives accurate results for an image, whereas k-means with OpenCV c++ does not give the same result, even though parameters are same. Is there any difference in implementation?

Answer 1

Matlab's kmeans function uses kmeans++ by default, and kmeans++ can be activated in cv::kmeans through the KMEANS_PP_CENTERS flag.

Answer 2

Wikipedia says that method randomly chooses k-observations from the data set and uses these as the initial means. And the result may depend on the initial clusters

That's why results may differ even within OpenCV methods execution, especially for Matlab