一种基于自适应单位链接PCNN的图像融合方法

摘要

A novel image fusion algorithm using adaptive unit-linking pulse coupled neural networks( ULPCNN) is put forward based on sufficient consideration of image local information. Firstly, ULPCNN threshold function is improved, the local discrete coefficient of each pixel is regarded as its interconnection strength, and then the null interconnection and the adaptive interconnection ULPCNN are formed. Secondly, each source image is processed by parallel ULPCNN, the nonlinear mapped time matrix can be obtained, which can represent the characteristic of the single pixel, and reflect the pixel information of the neighbourhood region in an image. Finally, through adaptive characteristic statistical judgment of each pixel and the neighbour local character in time matrix, corresponding fusion processing to source image can be established. Theoretical analysis and experimental simulations show that the proposed algorithm greatly overcomes the difficulty in adjusting parameters in PCNN, enhances image fusion performance with PCNN automatically. The abundant characteristic information is better concentrated in fused image. Compared with methods of PCA and Laplacian pyramid, this method presents higher detail average gradient, better visual effect and good fusion qualities.%在充分考虑图像局部信息的基础上,提出了一种基于自适应Unit-Linking PCNN赋时矩阵的图像融合算法.首先对ULPCNN阈值函数进行修正改进,并以每个像素的局部离散系数作为其链接强度,形成无连接和自适应连接ULPCNN;其次对各源图像并行进行ULPCNN处理,得到既能体现图像中单个像素特征,又能反映其邻域像素信息的非线性映射赋时矩阵；最后通过对赋时矩阵中诸像素及其邻域局部特征进行自适应统计判断,从而确定对源图像进行相关融合处理.理论分析和实验仿真表明,本方法极大地降低了PCNN参数多且设定难的问题,自动地提高了PCNN对图像融合处理的性能,融合图像较好地集中了源图像的丰富特征信息,融合细节清晰,视觉效果较好,融合质量优于主分量分析及Laplacian金字塔方法.