摘要
针对传统Hu氏不变矩易受摄像头径向畸变影响造成水下目标识别率低的问题,提出一种基于改进的Hu氏不变矩提取形状特征的方法,该方法依据摄像头的径向畸变模型重新恢复目标像素坐标与其灰度值的映射关系,构造出新的具有平移、缩放和旋转不变形的形状特征向量.同时为消除形状特征向量信息间的冗余问题,根据相关向量线性组合不改变向量自身性质的特点,提出一种基于相似度量准则进行对形状特征向量降维处理的方法.最后针对直接组合向量使得各特征权重一致而造成的识别率低的问题,给出了一种线性加权求和进行形状特征和纹理特征融合的方法.水下目标识别实验结果表明,该识别方法能够克服水下图像失真、信息冗余等不利因素,有效提高水下目标识别准确率.
According to the problem of low accuracy of underwater target recognition by traditional Hu＇s invariant moments,which is easily affected by the radial distortion caused by a camera,this paper put forward modified Hu＇s invariant moments and extracted target shape features.The mapping relationship of the target pixel coordinates and its gray value was recovered by the distortion model.The new shape feature vectors were reconstructed,after which they had the translation,scaling,and rotation invariant characteristics.To avoid the redundant problem between the information of the shape eigenvector,this paper proposed a method for reducing dimension processing based on similarity metric criterion.This method was based on the fact that the linear combination of relevant eigenvectors does not change the characteristics of its own property.Aiming at a low recognition rate because of the consistent feature weight achieved by direct combination of vectors,a method for shape and texture feature fusion through a linear weighted sum was proposed.The underwater experiment of target recognition shows that the proposed methods can overcome disadvantages such as image distortion and information redundancy,and can improve the accuracy of target recognition efficiently.
出处
《哈尔滨工程大学学报》
EI
CAS
CSCD
北大核心
2011年第9期1190-1195,共6页
Journal of Harbin Engineering University
基金
国防基础科研基金资助项目
黑龙江自然科学基金资助项目(QC2009C02)
关键词
特征提取
Hu氏不变矩
纹理特征
目标识别
feature extraction
Hu＇s moment invariants
texture feature
object recognition
