一种基于小波神经网络的图像融合方法  

Research on Image Fusion Based on Wavelet Neural Network

作　　者：龚庆凯[1] 曾黄麟[1] 赵雪专[2] 

机构地区：[1]四川理工学院自动化与电子信息学院,四川自贡643000 [2]中国科学院成都计算机应用研究所,四川成都610041

出　　处：《成都大学学报（自然科学版）》2013年第2期152-155,共4页Journal of Chengdu University（Natural Science Edition）

基　　金：四川省科技厅应用基础研究专项课题(2011JY0051)资助项目

摘　　要：图像融合结合图像处理、信号处理、计算机和人工智能等相关技术.通过对多源图像数据信息的提取合成,从而获得同一场景目标较为准确全面的图像描述.神经网络具有强大的非线性映射逼近能力,将神经网络用来进行滤波融合,避免了传统滤波图像变模糊问题.通过小波神经网络自适应动量快速学习算法进行图像滤波融合,能从根本上避免局部最优,且加快收敛速度,具有很强的学习和泛化能力,也避免了网络结构盲目设计.仿真实验表明,用本方法实现的融合图像更加符合人的视觉特性.Image fusion is a combination of image processing, signal processing, computer and artificial in-teUigence. Through the extraction and synthesis of information from multiple images of the same scene ob- jects, our research obtains the more accurate and comprehensive description of the image. Neural network has the strong approximation capability of nonlinear mapping and is used for filtering and fusion to avoid the blurring problem by traditionally filtering image. The wavelet neural network makes use of self-adaption momentum fast training algorithm for image fusion, which can effectively avoid local optimization and in- crease the convergence rate. The wavelet neural network has strong learning and generalization ability and avoids the blindness of network design. The experimental simulation shows that the image after fusion is more consistent with human visual characteristics.

关 键 词：图像融合 小波神经网络 小波变换 Morlet 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]