An Image Retrieval Method Based on a New Improved K-means Algorithm

According to the shortcomings of instability and low efficiency for K-means based image retrieval algorithm,this paper proposed an improved method to determine the initial class center of K-means clustering algorithm.Firstly,it uses HSV color space to calculate the feature vectors of sample image and all the images in the image database,and take the sample image's feature vector as the first initial cluster center,then calculate the distances of feature vector between all the images in the image database and sample image,and take the feature vector which has biggest distance as the second initial cluster center,and take the remaining feature vector of the most far away from the two initial cluster centers as the third initial cluster center,and determine the other initial cluster centers in the same way.Finally,the clustering are done according to the initial cluster centers to finish image retrieval.Experimental results suggest the validity of the algorithm.