文献类型：期刊文献

中文题名：一种高效的图像显著性检测算法

英文题名：An Efficient Algorithm for Saliency Detection

作者：范涛[1];朱煜[1]

机构：[1]华东理工大学信息科学与工程学院,上海200237

年份：2017

卷号：36

期号：10

起止页码：13

中文期刊名：实验室研究与探索

外文期刊名：Research and Exploration In Laboratory

收录：CSTPCD;;北大核心:【北大核心2014】；

基金：国家自然科学基金资助项目(61370174)

语种：中文

中文关键词：显著性检测;超像素分割;全局颜色对比;自动细胞机;改进的粒子群优化算法

外文关键词：saliency detection ; superpixel segmentation ; global color distinction ; cellular automata; particleswarm optimization

摘要：在人眼视觉特性的基础上,提出了一种高效的图像显著性检测方法。首先通过六边形简单线性可迭代聚类(HSLIC)对图像进行预处理,获得六边形的超像素块;再利用马氏距离定义显著块和背景种子块之间的距离,生成基于距离加权的全局颜色对比(GCD)初始显著图;然后引入自动细胞机模型对显著图进行优化。为进一步获取精确的显著性区域,提出一种改进的粒子群优化算法(NPSO)对显著图进行分割。所提出的算法在MSRA-5000和ECSSD数据库进行测试及比对分析。实验的结果表明,提取的显著图效果优异。
Saliency detection is the process of simulating the human eye to obtain image information, and is widely used in the field of computer vision. Based on the characteristics of human visual system, this paper presents an efficient method for image saliency detection. Hexagon Simple Linear Iteration Clustering （HSLIC） was used for pre-processing to get hexagonal pixel blocks. Then, the distances between the salient patches and the background seeds were calculated by Mahalanob, and the rough saliency map of global color distinction was obtained based on distance weighting. Next, cellular automata model was applied to optimize the saliency map. To obtain more accurate saliency map, we proposed a method improved by novel PSO algorithm to segment rough saliency map and get a better saliency region. We tested the proposed method on two standard datasets, MSRA-5000 and ECSSD. Experimental results show that the effect of saliency map is better than the state-of-the-art methods.