摘要
为了提高图像识别性能,采用核典型相关分析(KCCA)和支持向量机(SVM)相结合的方法进行图像识别。从不同维度获取图像文本特征,并采用KCCA方法对图像不同维度特征进行相关分析。将相关系数高的特征进行有效优化,从而保存差异度高的图像判别特征。对经过KCCA后的图像特征进行SVM判别,获得图像识别结果。实验证明,通过选择合适的KCCA核函数,对图像特征进行相关分析,并提取全面有效的图像判别特征,既降低了图像冗余特征,又保存了图像识别特征的全面性。与SVM、神经网络(NN)、卷积神经网络(CNN)相比,KCCA-SVM算法能够获得更高的图像识别准确率。
In order to improve the performance of image recognition,the combination of kernel canonical association analysis(KCCA)and support vector machine(SVM)is used for image recognition.Image text features are obtained from different dimensions,and then the KCCA method is used to perform a correlation analysis on the features of different dimensions of an image.Features with high correlation coefficient are effectively optimized,and the image discrimination features with high degree of difference are saved.The image features through KCCA are discriminated by SVM to obtain the image recognition results.Experiments show that by selecting an appropriate KCCA kernel function,correlation analysis of image features and extracting comprehensive and effective image discrimination features,it not only reduced the image redundancy features,but also preserved the comprehensiveness of participating image recognition features.Compared with SVM,neural network(NN),convolutional neural network(CNN)algorithms,KCCA-SVM algorithm can obtain higher image recognition accuracy.
作者
潘惠苹
任艳
徐春
Pan Huiping;Ren Yan;Xu Chun(School of Computer Science,Guangdong Business and Technology University,Zhaoqing 526000,China;Center for International Education,Philippine Christian University,Manila 1004,Philippine;School of Information Management,Xinjiang University of Finance and Economics,Urumqi 830012,China)
出处
《南京理工大学学报》
CAS
CSCD
北大核心
2022年第3期284-290,共7页
Journal of Nanjing University of Science and Technology
基金
国家自然科学基金(61966033)
广东省教育厅普通高校重点领域专项(2020ZDZX3105)。
关键词
核典型相关分析
支持向量机
图像识别
神经网络
卷积神经网络
kernel canonical correlation analysis
support vector machine
image recognition
neural network
convolutional neural network
