RAIC: Robust Adaptive Image Clustering

Abstract

Superpixels are becoming increasingly popular with the advent of new semantic segmentation and artificial vision techniques. As these techniques improve, there is a need for superpixel methods that can capture finer details. In this paper, we propose a new superpixel method called Robust Adaptive Image Clustering (RAIC), which, in addition to having the capability of capturing fine details, also offers the following advantages compared to other methods: high quality clustering, adaptive seed placement, compactness, robustness to noise, computational efficiency, and low algorithmic complexity. In addition, our algorithm has the tremendous advantage of being able to be written in the form of a cellular automaton. Cell automata are easy to parallelize, even massively, allowing for large performance gains when implemented for multithreaded CPU or GPU environments. Experimental results show that RAIC outperforms state-of-the-art superpixel segmentation algorithms. Furthermore, the results of the quantitative evaluation confirm the validity of the qualitative visual comparison of the superpixel image reconstructions.