摘要
研究了对完成计算机视觉任务有重要作用的视觉显著性检测,考虑到单纯依靠对比度计算进行显著性检测具有一定的局限性,提出了一种结合基元对比度与边界先验信息的显著性区域检测算法。该算法通过Mean-Shift分割构造图像基元结构,以图像基元为基础,利用图像颜色和亮度两种特征获得基元对比度显著图,再利用图像边界先验条件得到边界显著图;为了突出显著性目标,采用一种新的融合方式将以上检测结果进行融合,最后对显著图像进行多尺度增强操作,以获得更加高质量的显著性图。在国际公开数据集上的实验表明,该算法与现有的较成熟的方法相比,基本符合人眼的主观判断,具有较高的精度召回率。
The visual saliency detection, which plays an important role in completing computer vision and it was found that the detection relying on contrast calculation simply has its limitations, so tection algorithm based on the information of element contrast and boundary prior was proposed tasks, was studies, a salient region de- . Based on the ele- ment obtained by the Mean-Shift partition, the algorithm uses two kinds of characteristics of image color and lumi- nance to get the contrast saliency map. Then receives the boundary saliency map by taking advantage of the image boundary prior condition. It uses a new fusion method to fuse the above results. Lastly, the final saliency map with high quality is got by fusing the preliminary saliency maps of different scales. The results of the experiment on one of the largest publicly available data sets show that this algorithm can quickly extract salient regions consistent with human visual perception, and obtain the higher precision and better recall rate.
出处
《高技术通讯》
CAS
CSCD
北大核心
2015年第2期163-170,共8页
Chinese High Technology Letters
基金
国家自然科学基金(4020471)
河北省科技支撑计划(13211801D)
燕山大学博士基金(B540)资助项目
关键词
显著性区域
基元对比度
边界先验
特征融合
多尺度
salient region, element contrast, boundary prior, feature fusion, multi-level
