基于径向基函数网络的SFS算法研究

分析了现有从明暗恢复形状(SFS)的几种方法普遍存在对恢复的形状的连续性和光滑性的缺点,提出了一种基于径向基函数网络模型进行从明暗恢复形状的新算法。该算法先采用网络构造一个曲面方程,再利用反射函数作为约束条件,通过调整权因子和径向基函数中心和宽度对网络进行自学习,得到一个满意的曲面方程。理论和实验证明,该算法在恢复形状的准确性和曲面的光滑性,连续性上有较大改进。     

基于模糊调整的径向基函数网络图像恢复算法

为了解决传统图像恢复中存在的建模难的问题,提出了一种基于RBF神经网络的图像恢复算法,该算法利用RBF神经网络的非线性映射能力和适应性,通过记录退化过程的逆过程来恢复图像。首先改进RBF网络中心参数的确定过程,提出基于模糊调整的中心参数学习算法,然后用模糊调整后的网络进行图像恢复。仿真结果表明,改进的RBF网络可对典型退化图像进行令人满意的恢复。     

基于动态递归径向基函数神经网络的彩色图像恢复

将一种动态递归神经网络完成的最近邻分类器(NNC)应用于彩色图像恢复。采用多层感知器(WLP)与径向基函数(RBF)网络相结合的网络结构，把原型模式的显式表示作为网络参数，可自由扩大或删除原型模式，具有自适应特性。用这种模型实现的动态NNC去除了传统NNC中的比较运算,一定程度上降低了运算复杂度。试验结果表明该方法对于含有不同程度噪声的彩色图像恢复效果良好。     

基于径向基函数神经网络的非线性模型辨识

从径向基函数（RBF）神经网络原理分析出发,提出了一种基于RBF神经网络学习算法,用于对非线性对象模型的拟合与辩识,并将此方法用于实际非线性模型的学习与辩识。结果表明,基于RBF的神经网络可快速完成对样本的学习与拟合,对具有连续特性的线性与非线性模型,具有快速实时的学习速度和优良的学习性能。     

基于径向基函数模型的非线性预测控制策略研究

提出了一种基于径向基函数的非线性过程预测控制策略.首先,开发过程径基函数网络模型:根据过程特性,选择模型阶次、径基函数类型,用K均值法确定基函数中心位置,统计F检验确定基函数中心的数目,迭代最小二乘法确定径基函数网络权系数.然后,利用网络模型抽取非线性预测控制器(NLPC)特征样本训练构造径基函数网络预测控制器(RBFPC).仿真结果表明,与NLPC比较,由于RBFPC不必在线解非线性最优化问题,易于在线快速实施;与PI控制器比较,RBFPC具有更好的跟踪设定值性能和抗干扰性能.     

基于径向基函数神经网络的信用评分模型研究

对信用风险、信用评分进行了分析,在综合分析国内外企业信用评分指标体系的基础上,结合我国企业信用评分的特点,建立了适合我国企业信用评价的指标体系。结合国内外相关研究的现状与进展,及信用评分本身所具有的特点,建立了基于径向基函数神经网络的信用评分模型,利用现有数据分别进行判别和分析,研究其计算结果与实际情况的差距,然后使用改进的RBFNN学习算法,对径向基函数神经网络进行了学习训练,得到了令人满意的评价结果。利用该模型建立的评分系统具有进一步研究和推广应用的价值。     

基于径向基函数神经网络的信用评分模型研究

对信用风险、信用评分进行了分析，在综合分析国内外企业信用评分指标体系的基础上，结合我国企业信用评分的特点，建立了适合我国企业信用评价的指标体系。结合国内外相关研究的现状与进展，及信用评分本身所具有的特点，建立了基于径向基函数神经网络的信用评分模型，利用现有数据分别进行判别和分析，研究其计算结果与实际情况的差距，然后使用改进的RBFNN学习算法，对径向基函数神经网络进行了学习训练，得到了令人满意的评价结果。利用该模型建立的评分系统具有进一步研究和推广应用的价值。     

基于混沌映射的非径向对称基函数的神经网络模型

为提高神经网络模型的预测精度,构建了非径向对称基函数神经网络模型结构。为确定非径向对称基函数神经网络模型参数,采用Ulam-von Neumann映射规则确定混沌变量,利用混沌变量的遍历性获得不同网络结构参数下的最优网络输出,以减少所构建网络模型的实际输出与期望输出的差值,并利用模型输出的误差变化率以决定是否增加新的隐层节点。给出基于混沌映射的非径向对称基函数的网络模型构建步骤。采用基于Mackey-Glass时滞微分方程的混沌时间序列预测问题验证该模型的预测精度,并同其他文献对该序列预测的精度以及所需隐层节点数作对比。比较结果表明,采用该设计模型具有对时间序列预测精度高且所需网络结构规模小等优点。     

基于径向基函数神经网络的二维零件图特征识别

文章研究用径向基函数神经网络识别二维零件图的形状特征信息。提出了零件图扩展属性矩阵的概念及求法,对AutoCADR2000进行了二次开发,完善了笔者开发的二维回转类零件图特征识别和提取系统,具有一定的理论和实用价值。     

Local shape approximation from shading

Shading can be used as an independent cue for exact shape recovery, or it can be used as a supplementary cue for shape interpolation between features whose depths are known from other cues. Exact shape cannot be inferred from a local analysis of shading. However, for shape interpolation a crude local approximation may be sufficient. This paper explores the limits of such local approximations that are easy to compute. In particular, the shape of shading is used to approximate the surface in areas of monotonic change of intensity. This analysis is accompanied by a method for computing the direction of a single-point light source from the shading on occluding contours. A qualitative classification of shape near shading singularities is also discussed.This work was performed at the Massachusetts Institute of Technology, Center for Biological Information Processing.     

Linear-nonlinear neuronal model for shape from shading

The goal of shape from shading (SFS) is to recover a relative depth map from the variations of image intensity associated to changes in surface shape. There have been very few attempts at developing biologically plausible solutions to this problem, and a sound neurophysiological basis is still missing. Here we present a biologically inspired approach to SFS, formulated in terms of the well-known linear-nonlinear model of neuronal responses. Without resorting to the image irradiance equation, which is at the heart of the traditional SFS algorithms, we submit the input image to a linear filter followed by nonlinear transformations modelled on the tuning curves of the disparity-selective binocular neurons. This yields plausible shape estimates, without requiring information regarding surface reflectance or illumination.     

A Green's function approach to shape from shading

We present a new algorithm for shape from shading, inspired on the recently introduced disparity-based approach to photometric stereo (DBPS). Assuming that the single input image will be matched to a second image through a uniform disparity field, we construct an irradiance conservation equation and solve it for the matching image, via Green's function. When a linear expansion of the reflectance map is considered, this leads to a closed-form approximate expression for the surface function, whose parameters can be estimated via a structure-from-motion approach already used for DBPS.     

结合小波变换的SFS最小化算法研究

提出了一种结合小波变换的从明暗恢复三维形状的最小化方法,该算法利用图像小波变换各个子频段的不同频率特性和方向特性,分别采用不同的算法重构。在图像被分解后的低频区域采用小波直接提取自然条纹相位,高频区域使用最小化方法重构。这种方法在低频区域避免了SFS最小化方法的假设条件,而高频区域又发挥了它对细节部分重构的优势。实验结果表明该算法比单独使用最小化方法的误差要小。     

一种结构自适应的径向基函数神经网络

提出了一种新的结构自适应的径向基函数(RBF)神经网络模型。在该模型中,自组织映射(SOM)神经网络作为聚类网络,采用无监督学习算法对输入样本进行自组织分类,并将分类中心及其对应的权值向量传递给RBF神经网络,分别作为径向基函数的中心和相应的权值向量;RBF神经网络作为基础网络,采用高斯函数实现输入层到隐层的非线性映射,输出层则采用有监督学习算法训练网络的权值,从而实现输入层到输出层的非线性映射。通过对字母数据集进行仿真,表明该网络具有较好的性能。     

RBF神经网络的行车路径代价函数建模

行车路线优化是城市智能交通系统的研究热点之一,对整个交通系统的优化起着重要作用.分析了影响行车时间的各种因素,结合图论中最短路径算法,建立了基于RBF神经网络的路径代价函数模型.基于该函数模型,可以计算出交通图中任意给定两地间的时间最优路径.将该模型应用于实际路况进行有效性验证,得到了有实用价值的结果,说明了该模型的正确性和有效性.     

Existence and Uniqueness in Shape from Shading

For the image of a smooth surface object fully contained within the field of view and illuminated in and arbitrary direction,this paper discusses the existence and uniqueness o the conditions for solving a shape-from-shading problem under the conditions that the Fourier series expansion of the image intensity contains only zero and first order terms in a polar coordinate system.Three theorems are established,one for the existence and two for the uniqueness of z-axis symmetric shape from shading.     

基于高阶LLF和WENO算法的透视SFS

透视投影下从明暗恢复形状(SFS)问题,通常通过结合静态HamiltonJacobi(HJ)方程和快速扫描方法来求解。为进一步优化静态HJ方程的求解精度,改善透视SFS的恢复结果,采用了高阶局部LaxFriedrichs(LLF)通量分裂格式,以提高待求量的偏导数的精度;同时采用了改进的加权本质无振荡(WENO)格式,使得算法只计算整格点值,并且利用修正的光滑因子得到比WENO更高的精度。对合成图像和实际图像的实验结果表明,可以有效提高透视投影SFS问题的恢复精度。     

Spherical surface parameterization for perspective shape from shading

We propose a new mathematical formulation for perspective shape from shading (PSFS) problems. Our approach is based on representing the unknown surface as a spherical surface, expressed by Euclidean distance to the optical centre, as opposed to the traditional representation by distance from the image plane. We show that our parameterization is better suited for perspective camera models and results in simpler models and equations for classical PSFS problems with a light source in the optical centre. The unknown distance field satisfies a simple isotropic Eikonal equation on the unit sphere in the case of a Lambertian surface reflection model. This is in contrast to previous methods with depth field parameterization, which result in anisotropic equations. Adding light attenuation to the model, we show that the distance field satisfies an Eikonal type of equation with a zero order term. We show how both Eikonal equations can be approximated by very efficient Fast Marching methods. A number of numerical tests and examples are provided to demonstrate our approach, and to compare with previous work. Our results indicate competitive accuracy and computational time that are several orders of magnitude faster than state-of-the-art iterative algorithms. A preliminary investigation indicates that our method could be used in more general PSFS problems.