基于先验模型的全局线索选择感知编组算法  

Perceptual Organization Algorithm with Global Cue Selection Based on Prior Model

作　　者：周磊[1,2] 华钢[1] 徐钊[1] 徐冬梅[1,3] 

机构地区：[1]中国矿业大学信息与电气工程学院,江苏徐州221008 [2]悉尼大学信息技术学院 [3]杜伊斯堡-埃森大学自动控制与复杂系统研究所,德国杜伊斯堡47057

出　　处：《电子学报》2011年第8期1805-1810,共6页Acta Electronica Sinica

基　　金：国家自然科学基金(No.50534050)

摘　　要：不同的感知编组(perceptual organization,PO)算法针对不同的全局线索,在使用中必须由人首先判定目标所满足的全局线索,之后选择相应的编组算法进行计算.本文提出基于先验模型的全局线索选择感知编组算法,可以从待选的多个全局线索中挑选出概率意义下最可能的线索作为编组依据.先验模型将场景的统计特性作为先验知识,以广义拉普拉斯分布作为样本的估计分布,通过后验概率和线索概率得到归一化信息量,以表达不同全局线索在编组过程中的重要程度.本文同时提出了编组种子的优先级排序算法,以加快计算速度.最后,以煤矿监控场景为例,说明了算法的计算过程,实验结果验证了算法的有效性.Different perceptual organization（PO） algorithms focus on different global cues,so people have to firstly judge the object＇s global cue from the image,and then apply the corresponding PO algorithm.This paper proposed a PO algorithm with global cue selection based on prior model to automatically select the global cue with biggest probability.The prior model firstly obtained the prior knowledge and the sample distribution respectively from the statistical quantity of specific field and the generalized Laplacian distribution,then expressed the importance of each kind of global cues as the normalized information which was calculated by the posterior probability and the cue probability.To accelerate the computation,the grouping seed ranking method was incorporated.Finally,application of the algorithm in coal-mine field is analyzed.The results show this algorithm is efficient.

关 键 词：多线索选择 感知编组 先验模型 编组种子提取 编组种子排序 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]