Meaningful Region Extraction Based on Three-Stage Unsupervised Segmentation Algorithm

From a theoretical standpoint, meaningful region segmentation based only on gray level or color usually presents over segmentation or non-continuous regions. In view of this, we adopt a number of classical powerful algorithms (mean shift clustering, edge detection and region growing) to extract the meaningful regions adds spatial information. These algorithms are subjectively connected together and impact the results each other. The experiments indicate that the proposed method can avoid over-segmentation phenomenon and the results can be easily accepted by human eyes. Experimental results are superior to that of kmeans clustering method in both real-time performance and image segmentation performance. Finally, we achieved a new procedure to extract meaningful regions by clicking some place of a color image. It possesses a good application prospect and owns an effective real-time performance.