结合相位一致性和脉冲耦合神经网络的图像检索  

Image retrieval combining phase congruency and PCNN

作　　者：吴骏[1] 刘健[1] 肖志涛[1] 张芳[1] 耿磊[1] 

机构地区：[1]天津工业大学电子与信息工程学院,天津300387

出　　处：《天津工业大学学报》2014年第3期48-54,共7页Journal of Tiangong University

基　　金：国家自然科学基金(61102150);天津市科技支撑计划重点项目(12ZCZDGX02100);天津市高等学校科技发展基金(20120805)

摘　　要：为了有效地提取图像特征以提高图像检索性能,将具有生物视觉特性的脉冲耦合神经网络(PCNN)和符合人类视觉特性的相位一致性(PC)相结合,提出一种图像检索新方法.首先基于相位一致性,结合非极大值抑制和自适应双阈值法提取出图像的边缘特征,并获取边缘颜色直方图特征;然后对简化PCNN模型进行改进,针对PCNN神经元的链接强度通常为常数的不足,根据相位一致性自适应地调整神经元的链接强度,再利用改进PCNN模型提取图像的特征.最后综合运用基于PCNN的特征和基于相位一致性的特征进行图像检索.实验结果表明:该方法具有颜色和形状的鉴别能力,能获得较好的查准率和查全率.Inspired by biologic visual feature, a new algorithm for image retrieval using pulse coupled neutral network andphase congruency is proposed to effectively extract the features to improve the performance of image retrieval.Firstly, image edge feature is extracted by phase congruency, combing non-maximum suppression and auto-adaptive double-threshold. The edge color histogram is obtained by the image edge feature. Secondly, improvingthe simplified version of PCNN, the linking strength of each PCNN neuron is usually a constant, in order to over-come the limitation. Phase congruency is chosen to adjust its linking strengths according to image features adap-tively. Then image segmentation and feature extraction are carried out using the improved PCNN model. Finally,image retrieval is achieved by the features of PCNN and phase congruency. The experiment results show that themethod has the discrimination power against color and shape features, and has higher precision and recall.

关 键 词：相位一致性 脉冲耦合神经网络 边缘颜色直方图 图像检索 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]