Towards a Hybrid Approach of K-Means and Density-Based Spatial Clustering of Applications with Noise for Image Segmentation

Chun Guan, Kevin Kam Fung Yuen, Qi Chen

Abstract: Image segmentation is the process to divide a digital image into a number of regions for further analysis in the area of computer vision. Color images can be segmented by applying various clustering algorithms such as DBSCAN, which can identify the arbitrary shaped clusters. The drawback of DBSCAN is the high computational complexity whilst the sizes of image datasets are normally very large. This paper proposes a hybrid method of K-means and DBSCAN (Kmeans-DBSCAN) for image segmentation. K-means is the common partition-based clustering approach to reduce the size of image dataset. Four benchmarking image segmentation cases are used for evaluating the usability of proposed Kmeans-DBSCAN method.