多分辨率最小障碍与梯度融合显著性检测算法  

Multiresolution minimal barrier and gradient fusion saliency detection algorithm

作　　者：邵元 夏士雄[1] 姚睿[1] 牛强[1] 

机构地区：[1]中国矿业大学计算机学院,江苏徐州221116

出　　处：《计算机工程与应用》2018年第4期17-24,共8页Computer Engineering and Applications

基　　金：国家自然科学基金(No.61772530;No.61402483;No.61472267);江苏省自然科学基金(No.BK20171192);国家博士后基金(No.2016T90524;No.2014M551696)

摘　　要：自然图像的显著性区域一般处于图像中心,显著性区域可以通过计算区域与边界的距离得到,基于以上现象,产生了很多准确性较高或者运行速度很快的算法。但是要兼顾性能与效率,实现实时检测中需要的又快、又准确的显著性检测算法,仍需进一步研究。针对上述问题,提出了一种基于多分辨率最小障碍与梯度融合的显著性检测算法。通过多重采样形成多分辨率图像,引入改进后的最小障碍显著性检测算法对处理后的图像进行显著性检测;在算法运行过程中,对算法结果与背景线索图像进行梯度分析,将这两张图像进行融合,改善显著性区域模糊问题。经过在多个数据集上的实验验证,该算法能保证正确率在90%以上的情况下,检测速率达到75 f/s。The saliency region of natural image is generally located in the center. The saliency region can be obtained by calculating the distance between the region and the boundary. Based on the above phenomena, many algorithms with high accuracy or fast running speed are generated. However, for improving the efficiency and the performance of the algorithm in order to realize real time detection, it needs to do further research on obtainning faster and more accurate saliency detection. To handle the above problems, a saliency detection algorithm based on multi-resolution minimum barrier and gradient fusion is proposed. Firstly, the multi-resolution image is formed by multiple sampling, and the saliency detection algorithm based on the improved minimum barrier is introduced to caculate the saliency of the processed image. Secondly, during the operation of the algorithm, gradient analysis is performed on the results of the algorithm and the background image. The two images are fused to improve the problem of the fuzzy region. After experimenting on multiple datasets, the algorithm can achieve the detection rate of 75 f/s while the correctness is above 90%.

关 键 词：多分辨率 最小障碍算法 梯度融合 

分 类 号：TP302[自动化与计算机技术—计算机系统结构]