Design a effective parallel structure clustering approach for adaptive RGB image segmentation

Abstract

In a pattern recognition and image understanding applications, one of the most significant processes considered is the colour image segmentation. In this paper, an effective parallel structure clustering approach for adaptive unsupervised based bottom up red–green–blue (RGB) colour histogram search approach is proposed to achieve colour image segmentation. The RGB histogram is processed first by usinga double-scan procedure for the determination of important modes in every histogram. The bottom-up histogram search approach is used in the nest step to process the each mode for the RGB triplet formation. These triplets are used as the cluster centroids which cluster the pixels into regions and produce the resulting segmented image. In addition, a parallel processing structure is designed and implemented for the proposed scheme on agraphics processing unit (GPU) to reduce computational complexity. Experimental results show that the proposed algorithm outperforms state-of-the art algorithms both in terms of computational complexity and execution speed. For computational complexity, the proposed scheme running on a GPU provided19 and 15 times lower complexity than the proposed scheme running on a CPU and the L0-based scheme, respectively.