Artificial Fish Swarm Clustering Algorithm Based On Dynamic Niche And Its Application In Object Segmentation

Cluster analysis had always been a very important researcharea, and is widely used in signal processing, image processing, data mining, pattern recognition and other fields. Clustering is to given a certain set of design rules based on our practical problems which need to be divided into a number of categories which were always carried out in the condition with pre-knowledge. Currentlyalthough the traditional clustering algorithm such as Kmeans algorithm, FCM algorithm, and mean-shift algorithm had reached a very good performance on the clustering effect and operation efficiency, but a problem still exist to preset the number of category center. This paper focuses on intelligent clustering algorithm based on artificial fish, the goal is to achieve category centers and estimate the number of categories of clustering automatically and innovative research achievements include:(1) The proposed clustering algorithm based on artificial fish with dynamic niche, which can get the number of classes and centersfor clustering problem simultaneously. A more convenient way is used to describe an individualthat each individual behalves a regional center, through the dynamic operation allocate fishes into several niches and get the number of categories. The simulation and experiment shows the effectiveness of the proposed clustering algorithm. The algorithm was also applied to image segmentation which can effectively implement automatic segmentation of images.(2)Traditional behavior patterns of fishes has been improved by the introduction of artificial fish memory function which had changed the traditional artificial fish (AFSA)onlydepend on one direction to swim. By storing the position of its own memory fishescan swim under the guidance and direction of the two directions which can educing local oscillation phenomenon and speed up the convergence rate, thereby increasing the robustness and stability of the algorithm.(3) Finally, based on the proposed clustering algorithm, we combined the space divided weighted saliency detection algorithmachieved an natural image object segmentation algorithm, thus an unsupervised natural image object segmentation system was designed. The experiment reflects the very good results of the segmentation algorithm in visual effects and evaluation indication.