Robust and Accelerated Frequency-Tuned Salient Object Detection via Color Quantization

Abstract

Frequency-tuned (FT) saliency detector has been widely used in various image processing and computer vision applications. However, removing high frequency noise and textures by a Difference of Gaussians (DoG) filter is still time-consuming, which hinders real-time execution of FT. In this paper, an improved algorithm is proposed to tackle the obstacles to performance improvement. It exploits color quantization within natural images and substitute Gaussian filtering process. Experimental results demonstrate that the proposed method runs up to about two times as fast as FT while resulting in almost the same saliency map.