稀疏FRAME模型的感兴趣目标检测算法  

Object Detection Algorithm Based on Sparse FRAME Model

作　　者：杨建秀[1] 胡正平[2] 

机构地区：[1]山西大同大学物理与电子科学学院,山西大同037009 [2]燕山大学信息科学与工程学院,河北秦皇岛066004

出　　处：《四川兵工学报》2014年第12期107-111,共5页Journal of Sichuan Ordnance

基　　金：国家自然科学基金(61071199);河北省自然科学基金(F2008000891;F2010001297);中国博士后自然科学基金(20080440124);第二批中国博士后基金特别资助(200902356)

摘　　要：针对基于非齐次FRAME(Filters,Random filed,And Maximum Entropy)模型的目标检测算法在目标发生较大形变或存在阻挡等情况下的定位产生一定偏差,以及学习模型所需大量时间等各方面的研究,提出了一种稀疏FRAME模型的感兴趣目标检测算法。首先用共享稀疏编码方法对样本图像进行特征提取,由所选择的基函数构成可变形的稀疏FRAME模型;然后用交替的求和图及最大值图结构对测试图像进行匹配检测,实现目标定位。经多组实验结果表明,该算法不仅在一定程度上提高目标发生较大形变或存在阻挡等情况下的鲁棒性,而且节省了大量的样本训练时间。Object detection algorithm based on inhomogeneous FRAME model was difficult to solve the problem when target occurred large deformation or the presence of occlusion,and needed a lot of time on training model,so this paper presented a detection algorithm of object-of-interest based on sparse FRAME model to solve these problems. Firstly,the algorithm selected features via a shared sparse coding scheme,and formed the deformable sparse FRAME model by selecting shared basis functions. Secondly,the matching of the sparse FRAME template to a testing image could be accomplished by a cortex-like structure of recursive sum-max maps. The experimental results show that this method is robust when target occurred large deformation or the presence of occlusion,and saves a lot of training time.

关 键 词：目标检测 稀疏模型 共享编码 非齐次 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]