摘要
文中研究了无监督自下而上的显著性目标检测方法。基于显著性目标在自然图像中稀疏分布的这一先验性假设,提出了一种用低秩和稀疏表示进行显著性目标检测的方法。根据图像背景的先验分布,首先选取一个有效的背景字典来低秩表示图像的背景部分,进而更好地分离出显著性前景。由于人类视觉中心偏好可知,图像的边缘部分不易引起关注,故选取这些边缘部分作为背景先验来选取背景字典。与其他基于稀疏和低秩分解的显著性目标检测相比,文中选取的背景字典更简单有效,且能得到更好的显著性图。实验结果显示,该方法比主流的显著性检测方法得到的显著性图更令人满意。
This paper studies the unsupervised saliency detection method. Based on the assumption that salient objects are sparsely scattered over the image,we propose the sparse and low rank based saliency detection method.According to the priori distribution of image background,we select an effective background dictionary to represent the image background and thus separate the salient foreground. Compared with other saliency detection method the background dictionary in our method is simple and effective,and can help locate the salient objects accurately. Experimental results show that compared with other method a difference does exist in favor of ours.
出处
《电子科技》
2015年第2期112-115,共4页
Electronic Science and Technology
关键词
稀疏表示
超像素分割
显著性目标检测
sparse representation
sup-pixel segmentation
saliency detection
