K-Means clustering based on density for scene image classification

K-means clustering has been extremely popular in scene image classification. However, due to the random selection of initial cluster centers, the algorithm cannot always provide the most optimal results. In this paper, we develop a density-based k-means clustering. First, we calculate the density and distance for each feature vector. Then choose those features with high density and large distance as initial cluster centers. The remaining steps are the same with k-means. In order to evaluate our proposed algorithm, we have conducted several experiments on two scene image datasets: Fifteen Scene Categories dataset and UIUC Sports Event dataset. The results show that our proposed method has good repeatability. Compared with the traditional k-means clustering, it can achieve higher classification accuracy when applied in multiclass scene image classification.