Salient object detection by combining multiple color clustering
Kang Han Oh, Soo-Hyung Kim, Young‐Chul Kim · 2015
This paper presents a novel clustering-based approach for computing a salient object. The key idea of the proposed method is that saliency is detected by using multiple color models with different Gaussian filters to derive various segmentation results. The proposed method consists of two main processes: mean-shift based saliency (MS) and Bayesian based saliency (BS). First, three different models for the input image are created using different Gaussian filters. Then, the MS process categorizes all of the pixels, and the categorized results are utilized to extract saliency using centroid weight map (CWM) and foreground estimation (FE). For the BS method, saliency is detected in a similar manner, but the difference between MS and BS is that the BS categorizes all of the pixels using the prior knowledge from mean-shift results. In the experimental results, the scheme achieved superior detection accuracy in the MSRA-ASD benchmark database with both a higher precision and better recall than state-of-the-art saliency detection methods.