摘要
唐卡图像具有内容丰富、画面复杂、色彩表现层次多等特点,但部分图像存在破损残缺、清晰度不高的不足。为此,提出一种融合Hu矩和局部二进制模式的图像特征提取方法,提高目标图像的特征表现程度,并在此基础上给出一种改进的基于距离密度的K最近邻分类算法,该算法待测样本所属的类别与其周围邻近点的类别最相关,且距离越近,相关度越高。实验结果表明,与传统的k NN、神经网络和神经网络集成方法相比,该方法对唐卡图像中的法器对象具有更高的识别正确率,能有效实现唐卡图像中法器对象的分类识别。
Thangka image is different from landscape image or other art image. It has rich colors,irregular forms,nonrepresentational texture,plentiful content,complex composition,and overflowing with miniscule and elaborate details.According to the special characteristic,multi-feature based religious ritual implement recognition method in Thangka image is proposed. The Hu moments and Local Binary Pattern（ LBP） are combined to describe the feature of object image effectively,and a newalgorithm Distance Density k Nearest Neighbor（ DDk NN） classification algorithm is given,the main idea of this algorithm is that the class of test sample is related to the neighbor training samples,and the closer the distances,the higher the relevance becomes. Experimental results showthat,compared with traditional k NN,BP neural network and Integrate BP neural network,this algorithm is simple and effective to recognize for religious ritual implements in Thangka image.
出处
《计算机工程》
CAS
CSCD
北大核心
2016年第3期198-203,207,共7页
Computer Engineering
基金
国家自然科学基金资助项目(60875006
61162021)
西北民族大学科研创新团队基金资助项目
关键词
唐卡图像
图像识别
k最近邻分类算法
神经网络
局部二值模式
Thangka image
image recognition
k-Nearest Neighbor（kNN） classification algorithm
neural network
Local Binary Pattern（LBP）
