基于颜色差异计算的图像显著性检测算法研究  

Research on Image Saliency Detection Algorithm Based on Color Difference Calculation

作　　者：邵凯旋 童林 吴帮吕 杨文韬[2] 祝昆[2] SHAO Kaixuan;TONG Lin;WU Banglv;YANG Wentao;ZHU Kun(Yunnan University,Kunming 650500;Liupanshui Normal University,Liupanshui 553004)

机构地区：[1]云南大学,昆明650500 [2]六盘水师范学院,六盘水553004

出　　处：《计算机与数字工程》2022年第2期394-398,共5页Computer & Digital Engineering

基　　金：贵州省科技厅联合基金项目(编号:黔科合LH字[2014]7456号);六盘水师范学院硕士学位培育点(编号:LPSSYSSDPY201704);六盘水师范学院校级项目(编号:LSZDZY2018-03,LPSSYsyjxsfzx 201801)资助。

摘　　要：针对目前显著性算法普遍面临的背景噪声较多,准确性较低等问题,提出一种基于图像颜色对比的显著性检测算法。首先在Lab伪彩色空间域将图像进行SLIC超像素分割,依照图像的全局对比计算,确定前景预选区域。然后以该区域内超像素特征通道信息值作为参考值,计算各超像素显著值,得到初步显著图。最后构建视觉中心,考虑颜色权重信息对显著目标区域的影响,优化得到最终结果。采用当前通用数据集进行测试,并与其它六种主流算法进行对比,结果显示论文实验算法不仅最终结果图质量较优,还在多个客观评价指标上表现良好。Aiming at the problems that the current saliency algorithms generally face are more background noise and lower accuracy,a saliency detection algorithm based on image color contrast is proposed. First,the image is divided into SLIC superpixels in the Lab pseudo-color space domain,and the foreground preselected area is determined according to the global comparison calculation of the image. Then,using the superpixel feature channel information value in the area as a reference value,the superpixel saliency value is calculated to obtain a preliminary saliency map. Finally,a visual center is built,the influence of color weight information on the significant target area is considered,and it is optimized to obtain the final result. The current general data set is used for testing and compared with the other 6 mainstream algorithms. The results show that the experimental algorithm in this paper not only has better final result graph quality,but also performs well on multiple objective evaluation indicators.

关 键 词：显著图 前景预测 颜色权重 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]