Multi-class image annotation approach using particle swarm optimization

This paper presents an automatic image annotation approach for region labeling. The proposed approach is based on multi-class k-nearest neighbor, K-means, and particle swarm optimization algorithms for feature weighting, in conjunction with normalized cuts based image segmentation technique. This hybrid approach refines the output of multi-class classification that is based on the usage of k-nearest neighbor classifier for automatically labeling image regions from different classes. Each input image is segmented using the normalized cuts segmentation algorithm in order to subsequently create a descriptor for each segment. Particle swarm optimization algorithm is employed as a search strategy to identify an optimal feature subset. Experimental results and comparative performance evaluation, for results obtained from the proposed particle swarm optimization based approach and another support vector machine based approach presented in previous work, demonstrate that the proposed particle swarm optimization based approach outperforms the support vector machine based one, regarding annotation accuracy, for the used dataset.