基于HVS的多尺度显著性检测算法

摘要

为提高图像显著性检测的准确性,借鉴有关人类视觉系统的研究成果,提出了一种基于人类视觉系统(HVS)的多尺度显著性检测方法.该方法先将图像分割成小的图像片以获取图像的局部信息,然后采用PCA进行特征抽取,在得到的低维空间中计算图像片的差异.通过结合人类视觉系统和多尺度方法降低背景的显著度,提高显著性目标的显著值.实验结果表明,该方法在检测效果和抗噪能力等方面均可获得较为满意的结果.%This paper proposed a model for detecting salient regions in an image based on human visual system, and proposed multiple scales to improving the accuracy of salient detection. It divided original image into small image patches for gathering local information. Then it used the principal component analysis to extract features which used to calculate the dissimilarities among the image patches. To combine HVS and multiple scales, it could decrease the saliency of background and increase the saliency of salient objects in the original image. The results of experiments with many real images show the algorithm is effective and efficient.