Superpixel Clustering Based Segmentation Algorithm for Hyperspectral Image Classification

Abstract

Superpixel segmentation approaches have gained a lot of popularity in recent years. It divides the image into several semantic sub-regions to simplify the subsequent image processing tasks. The number of generated superpixels has significant effect on the classification results and is usually kept high to capture the underlying distinct characteristic features in the image. Inthis paper, Simple Linear Iterative Clustering (SLIC) algorithmis employed to generate an initial segmentation map. This map isfurther processed, by clustering the similar superpixels togetherusing the Density-based spatial clustering of applications withnoise (DBSCAN) algorithm and hence a final segmentation map is produced. Then a majority voting is performed between thefinal segmentation map and the pixel-wise classification map to generate the final classification map. The performance of the proposed algorithm is validated over the well-known Indian Pinesand Pavia University datasets.