摘要
针对大多数现有基于内容的图像检索方法的性能很大程度上依赖分类器的问题,提出了一种基于模糊隶属度融合神经网络的CBIR方法;首先,利用离散小波变换进行特征提取;然后,使用神经网络计算查询图像的类标签和模糊类隶属度;最后,利用简单与加权距离度量的组合在完整搜索空间中进行检索;在3个纹理类数目、方向和复杂度都不同的数据库上进行实验验证了所提方法的有效性,实验结果表明,相比其他几种较新的纹理图像检索方法,所提方法取得了更好的检索性能。
For the issue that performance of existing content--based image retrieval approaches depends very much on classifier, a CBIR approach based on fuzzy membership degree fusion with multilayer feed--forward neural network is proposed. Firstly, discrete wavelet transform is used to extract features. Then, neural network is used to calculate the label and fuzzy membership degree of query image. Finally, the combination of simple and weighted distance measure is used to do retrieval in complete search space. The effectiveness of proposed ap- proach has been verified by experiments on three databases with different texture class number, direction and complexity. Experimental results show that proposed approach has better performance than several advanced texture image retrieval approaches.
出处
《计算机测量与控制》
2015年第3期903-906,909,共5页
Computer Measurement &Control
基金
国家自然科学基金资助项目(71001005)
关键词
模糊隶属度
基于内容的图像检索
纹理图像
多层前馈神经网络
加权距离度量
fuzzy membership degree
content--based image retrieval
texture image
multilayer feed--forward neural network
weighted distance measure
