摘要
为了解决目前显著性检测中图像的纹理特征对其他特征(颜色、方向等)的影响问题,通过对该领域已有的算法和文章进行实验和比较,提出一个概念上明确和改进的基于对比度的显著性检测的直观算法。研究了将输入的图像用已改进的总变差模型来获取一个有效的主要结构,将提取到的主要结构图像分解为紧凑、感知、同质的元素,抽象出不必要的细节;进一步,基于这种抽象元素计算了两个对比度量,即颜色独特性和空间分布,通过高斯滤波得到两个对比度的显著图,将其进行融合。分析了融合后的显著图与目前比较经典的算法得到的显著图之间的改进之处。实验证实了去除纹理特征后提取到的显著图,有效地抑制了图像各特征之间的相互干扰,提高了显著图融合的精确度。
In order to solve the problem of the influence of the image texture features on other features(color,direction,etc.)in the current saliency detection,a contrast-based intuitive algorithm is proposed for saliency detection with the clear concept and improved performance by experimenting and comparing the existing algorithms and articles in the fields.The input image is extracted with an improved total variation model to obtain an effective main structure,which is decomposed into compact,perceptual,homogeneous elements,and abstracts unnecessary details.Furthermore,based on the abstract elements,two contrast quantities are computed,namely color uniqueness and spatial distribution,which are fused by Gaussian filtering to obtain a saliency map with the two contrasts.The improvement between the fused saliency map and the one obtained by the classical algorithms is analyzed.The experiment demonstrates that the saliency map without the texture features effectively suppresses the mutual interference among various image features,and improves the accuracy of the saliency map fusion.
作者
艾显丽
彭亚雄
陆安江
AI Xianli;PENG Yaxiong;LU Anjiang(Guizhou University,School of Big Data and Information Engineering,Guiyang 550025,China)
出处
《移动通信》
2019年第10期92-96,共5页
Mobile Communications
基金
贵州省科技重大专项:草海综合整治工程大数据系统集成与示范(黔科合重大专项字[2016]3022号)
关键词
显著性检测
纹理特征
对比度
saliency detection
textural property
contrast ratio