Abstract:

In order to overcome the initial sensitivity of the original spectral clustering,a novel image clustering method CIMO is presented based on the invasive seed optimization（IWO）.In this method,the optimal cluster number is dynamically determined by calculating the Peak Signal-to-Noise Ratios（PSNRs）,and a new evaluation function of clustering quality is redefined by employing the minimum quantity error,the minimum intra-instance and the maximum inter-instance.Moreover,the clustering centroids of image datasets are quickly and accurately located by simulating the natural behaviors of weed colonization.The proposed algorithm are then applied to several test benchmark images and are compared with the well-known methods such as k-Means,FCM and PSO via the clustering validation criterions.The results indicate that the proposed CIWO method is of higher clustering stability and better clustering quality.

Key words: pattern recognition, image clustering, invasive weed optimization, image analysis, optimization