基于证据理论融合多特征的物体识别算法  

Object recognition method based on multi-feature fusion of D-S evidence theory

作　　者：孙晋博[1] 余隋怀[1] 陈登凯[1] 

机构地区：[1]西北工业大学机电学院工业设计研究所,西安710072

出　　处：《计算机工程与应用》2015年第9期147-151,共5页Computer Engineering and Applications

基　　金：国家高技术研究发展计划(863)(No.2009AA093303);高等学校学科创新引智计划(No.B13044)

摘　　要：为了提高物体的识别正确率,提出一种基于证据理论融合多特征的物体识别算法。提取物体图像的颜色直方图和尺度不变特征,采用极限学习机建立相应的图像分类器,根据单一特征的识别结果构建概率分配函数,并采用证据理论对单一特征识别结果进行融合,得出物体的最终识别结果,采用多个图像数据库对算法有效性进行测试。测试结果表明,该算法不仅提高了物体的识别率,而且加快了物体识别的速度,具有一定的实际应用价值。In order to obtain better recognition results, a novel object recognition method based on multi-feature fusion of evidence theory is proposed. Color histogram and scale invariant feature transform features are extracted from object image, and extreme learning machine is used to establish the classifier;the recognition results of single feature are fused to obtain the last recognition results of object based on evidence theory;the performance of algorithm is tested by some image data. The result illustrates that the proposed algorithm has improved the recognition rate and speed, and it has some application vale.

关 键 词：物体识别 证据理论 极限学习机 尺度不变特征变换 颜色直方图 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]