项目名称: 视感知模型脉冲耦合神经网络的图像特征提取及应用研究
项目编号: No.61463052
项目类型: 地区科学基金项目
立项/批准年度: 2015
项目学科: 自动化技术、计算机技术
项目作者: 聂仁灿
作者单位: 云南大学
项目金额: 47万元
中文摘要: 出于寻求一种对外界变化具有强鲁棒性的图像特征提取方法的目的,并由于从脉冲耦合神经网络(PCNN)产生的二值图像序列和频图序列中提取到的一维振荡时间序列(OTS),对图像的几何变化和噪声具有良好适应性,本项目将研究基于PCNN的图像OTS特征自动提取方法。首先研究神经元的动力学机制,及各种动力学特性对OTS特征可鉴别的影响。进而通过研究分析神经元参数的动态范围,及这些参数对OTS特征可鉴别性的影响,研究提出适用的神经元参数的自动估计方法。再之利用二维图像转化为零维数据点的变换方法,研究从二值图像序列或频图序列中提取OTS特征的方法。同时利用完备的脉冲分类方法,研究图像的OTS特征分解方法,以提高OTS特征的可鉴别性。最后研究适合于单/多OTS特征的分类器设计,以在实际应用中实现OTS特征模式的良好分类。研究成果预期为实现强鲁棒性的图像特征提取提供一种全新的思路和有效的支撑技术。
中文关键词: 脉冲耦合神经网络;图像特征提取;动力学机制;参数估计;模式分类
英文摘要: To seek an image feature extraction method with strong robustness for the environment change, and because of one-dimensional oscillation time series (OTS) from binary image sequence and frequency chart sequence generated by pulse coupled neural network (PCNN) has a good adaptability for geometric changes and noise of image, this project will study image OTS feature automatic extraction methods based on PCNN. First we will study neuronic dynamics mechanism, and various kinds of dynamics characteristics influence on image OTS feature. And then, by analyzing the dynamic range of parameters of neurons, and these parameters influence on image OTS feature, we will study and put forward the applicable automatic estimation methods of the parameters of neurons. And using transformation methods of two-dimensional images translating into zero dimension data point, we will research methods from binary image sequence or frequency chart sequenceextracting OTS feature. Meanwhile using complete pulse classification methods, we will study the OTS characteristics decomposition methods of image to improve the identifiability of OTS feature. Finally we will research classifier design of single/multiple OTS feature to achieve good classification of OTS feature model in the practical application. The project's successful implementation is expected to realize the image feature extraction with strong robustness and provide a new thinking and effective technology support.
英文关键词: Pulse Coupled Neural Network;image features extraction;dynamic mechanism;parameters estimation;pattern classification