Standard and Genetic k-means Clustering Techniques in Image Segmentation
CISIM '07 Proceedings of the 6th International Conference on Computer Information Systems and Industrial Management Applications
A hierarchical approach to color image segmentation using homogeneity
IEEE Transactions on Image Processing
Segmentation by Fusion of Histogram-Based -Means Clusters in Different Color Spaces
IEEE Transactions on Image Processing
| Hi-index | 0.00 |
In the resource-constrained mobile computing environment, new challenges emerge for multimedia applications, e.g., how to implement highly computing-intensive multimedia processing algorithms on resource limited hardware to achieve real-time performance. To address this problem, we propose a novel K-Means based algorithm to improve the efficiency of image segmentation, which is the key step in many multimedia content analysis tasks. As a traditional clustering algorithm, K-Means has been widely adopted for image segmentation due to its simplicity and easy implementation. However, the classical K-Means suffers from running with a user-defined K value and the random selection of initial centroids which can degrade the clustering quality. Additionally, it is time-consuming when processing a large amount of data. Moreover, K-Means does not consider the spatial properties in images, leading to false segmentation boundaries. To solve the above problems, three strategies are employed in this work. Firstly, through analysis on chromatic histogram, dynamically adjusted K value and initial clustering centroids can be obtained automatically. Secondly, the number of iterations for clustering analysis has been reduced by adopting image pyramid. Finally, the spatial edge information contained in the pyramid is used to refine the clustering result. Experiments on Berkeley segmentation dataset and benchmark (BSDS) demonstrate that our proposed scheme has significant advantage over classical K-Means on processing speed, while producing comparable or even better segmentation quality.