Superpixel Segmentation using DBSCAN Clustering Algorithm

Abstract

Abstract

Superpixel division strategies are generally utilized as a pre-preparing venture to accelerate picture handling tasks. They assemble the pixels of a picture into homogeneous locales while attempting to regard existing shapes. In this paper, we propose a quick superpixels division calculation with contour adherence utilizing the density based spatial clustering of applications with noise (DBSCAN) algorithm. We implement a two-stage processing to decrease the computational costs of superpixel algorithms. First is clustering stage and the second is merging stage. At first, the DBSCAN algorithm with color similarity and geometric restrictions are used to cluster the pixels quickly, and then, small clusters are merged into superpixels by their neighborhood. It is done using a distance, which is defined by color and spatial features. We have defined a robust and simple distance function to obtain better superpixels. The experimental results show that our proposal outperforms the state-of-the-art superpixel segmentation methods in terms of both accuracy and efficiency.

Keywords: Superpixels, DBSCAN, segmentation, under segmentation error, boundary recall

Cite this Article

Partha Ghosh, Kalyani Mali, Sitansu K. Das Superpixel Segmentation using DBSCAN Clustering Algorithm. Journal of Artificial Intelligence Research & Advances. 2017; 4(3): 26–36p.