Nonparametric K-means clustering-based adaptive unsupervised colour image segmentation

Image segmentation focuses at highlighting region of interest within the image, by accumulation of pixels based on given properties. This task resembles to clustering, yet many standard clustering methods fail to meet the basic requirement of image segmentation, that is number of segments is rarely determined automatically. The proposed nonparametric K-means clustering (EAIS) overcomes this limitation and turns out to be particularly suitable for the task of image segmentation. In this paper, we propose a nonparametric K-means clustering approach (EAIS) that automatically and adaptively determines the initialization conditions, i.e. number of clusters, initial cluster centroids, and subsequently segments the image into suitable regions. The proposed approach comprises of five modules that includes deep image reconstruction, intra-histogram individual peak level detection, inter-histogram peak levels association, mutual consensus-oriented cluster seeds merging, and morphological reconstruction-driven spatial post-processing. Deep reconstruction performs image smoothing by reducing the variance and outliers in the colour channel distribution. The proposed approach utilizes image histograms-based global distribution to determine the optimal initialization condition for pixel clustering (image segmentation). Followed by dynamic and optimally devised cluster seeds merging for redundancy reduction and determination of adequate number of cluster seeds for K-means initialization. Finally, morphological reconstruction inducts spatial awareness in the clustered space and enhances the spatial consistency of cluster member's (pixels). Diverse experimental results on the BSDS500 benchmark validate that our proposed approach is robust to various natural scenarios and comparable to state-of-the-art methods regarding segmentation quality and computational efficiency.