摘要
针对传统目标显著性检测算法存在显著区域弱化、最显著的中心点被抑制、背景差对比度低等问题,提出一种新的整形目标显著性检测算法。算法首先利用灰度不一致算子作为局部处理手段,刻画图像局部纹理的非均匀性,使得最显著的中心点亮度提高;其次,利用改进的FT算法,建立一种新的全局量化方法,使得显著区域增强;再次,为了滤除孤立显著区的影响,算法提出一种空间权重表达法,对所提显著图进行线性处理,提高整体显著区与背景间的对比差。最后的仿真实验中,与FT、Itti等6种典型的目标显著性检测算法相比,该算法不仅具有更好的识别准确性和稳定性,而且所提算法的精确率和召回率等客观指标也具有较强的优势,从而表明该算法是切实可行的。
Aiming at the problems in traditional object saliency detection algorithms that the salient region is weakened,the most salient pixels are suppressed,and the contrast of background difference is low,etc.,we proposed a new integral form object saliency detection algorithm. First,the algorithm utilises the gray inconsistency operator as the means of local processing to depict the non-uniform of local texture,which makes the brightness of the most salient pixels be increased; Secondly,it uses the improved FT algorithm to set up a new global quantification approach,this makes the salient area be significantly enhanced; Thirdly,in order to filter out the influence of the isolated salient areas,the algorithm proposes a spatial weight expression method to carry out linear processing on the saliency map mentioned so as to improve the contrast between the integral salient area and the background. In final simulation experiments,compared with 6 typical object salient detection algorithms including FT,ITTI,etc.,the proposed algorithm not only has better recognition accuracy and stability,but also has higher advantage in objective indicators of precision rate and recall rate,therefore indicates that the proposed algorithm is feasible.
出处
《计算机应用与软件》
CSCD
2016年第11期204-207,共4页
Computer Applications and Software
基金
国家自然科学基金项目(61373089)
