摘要
为了提高显著图的精确度,能够准确地提取显著目标,提出了结合局部对比和全局稀有度的显著性检测算法。该方法通过两个方面来衡量显著性:局部对比和全局稀有度。采用多尺度高斯差分模拟"中央周边差"计算中心点与其周围点的各特征之间的差异来描述局部对比,采用多尺度高斯卷积图像的特征及特征方差的概率计算全局稀有度,将局部对比和全局稀有度融合和归一化得到最终综合显著图。实验结果表明,该方法能较好地检测图像中的显著区域,在突出图像中不同物体边缘的同时,也能突出图像中同质区域的显著性,说明了结合局部和全局的算法能够得到更好的显著图。
In order to improve the accuracy of saliency map and extract salient object more precisely, this paper proposed a algorithm combined local contrast and global rarities. The algorithm measured saliency by two measures: local contrast and global rarity. It simulated ' center-surround difference' with multi-scale difference of Gaussian to calculate the difference be- tween features and surrounding features to get local contrast. By using the probability of features and features' variance of multi-scale Gaussian convoluted image, global rarity of each pixel could be described. It fused and normalized the local con- trast and global rarity to get the final integrated saliency map. Experimental results show that this method can detect salient re- gions in an image, it can not only highlight the edge of different object in images, but also highlight the saliency of homogene- ous region in image. It shows that model which combine local and global saliency can get better saliency map.
出处
《计算机应用研究》
CSCD
北大核心
2014年第9期2832-2835,2840,共5页
Application Research of Computers
基金
国家自然科学基金资助项目(61240059)