摘要
为了提高图像语义特征提取的精确度,克服目前大部分图像语义特征提取算法中,因图像特征提取不当,导致特征参数不能全面反映图像语义的问题,提出了一种基于典型相关分析(CCA)的特征融合的图像语义特征提取方法。该方法首先采用圆形对称邻域取代传统的矩形邻域的方法,对局部二值模式(LBP)纹理特征进行了改进,然后采用高维小样本下典型相关分析对可伸缩颜色描述算子的颜色特征和改进的LBP纹理特征进行特征融合。实验结果表明,所提出的方法明显提高了图像语义特征提取的精确度,能有效地建立图像的低层特征与语义特征间的一致性。
For the purpose of have a better accuracy of image semantic feature extraction,overcome the problem that in most image semantic feature extraction algorithms,due to the improper extraction of image semantic feature,lead to the problem of feature parameters can not fully reflect the image semantic,this paper proposed an image semantic extraction algorithm based on canonical correlation analysis and feature fusion.In the proposed method,using the circular symmetric neighborhood,instead of the traditional method of rectangular neighborhood firstly,improved the local binary patterns（LBP） texture feature descriptor.Then,the work was feature fusion between the scalable color descriptor color feature and improved LBP texture feature using canonical correlation analysis under high dimension small sample.Experimental results show that the proposed method significantly improves the accuracy of image semantic feature extraction,creates the consistency between low-level features and high-level semantic effectively.
出处
《计算机应用研究》
CSCD
北大核心
2012年第5期1938-1942,共5页
Application Research of Computers
基金
安徽省自然科学基金资助项目(11040606M134)
安徽省高校自然科学基金重点资助项目(KJ2009A150)
关键词
图像语义
典型相关分析
局部二值模式
特征参数
特征融合
image semantic
canonical correlation analysis
local binary patterns
feature parameters
feature fusion
