基于多尺度变换的PCNN和FOA图像融合  

Image fusion based on PCNN and FOA of multi-scale transform

作　　者：李美丽[1] 

机构地区：[1]西安石油大学理学院,陕西西安710065

出　　处：《光电子．激光》2016年第7期767-772,共6页Journal of Optoelectronics·Laser

基　　金：国家自然科学基金(61472471);陕西省教育厅专项科研计划(15JK1573);西安石油大学博士创新基金(ys29031208)资助项目

摘　　要：为解决传统的图像融合算法融合质量不高的问题,提出一种采用果蝇优化算法(FOA)自适应地选取脉冲耦合神经网络(PCNN)的4个参数并将其与多尺度变换相结合的图像融合方法。首先利用4种多尺度变换对待融合图像进行分解,对得到的低频分量采用PCNN和FOA的融合规则进行处理,对高频分量采用绝对值最大的原则进行系数选择,最后通过逆变换得到融合后的图像。实验结果表明,与常用的融合规则对比,在主观效果上,本文融合规则能够更有效地保留源图像中的细节信息,提高融合图像的质量;在客观指标上,本文方法的融合图像在互信息(MI)、边缘保持度Q^(AB/F)、熵(entropy)、平均结构相似度(MSSIM)以及标准差(SD)等客观评价指标上更为优越。Aiming at the fusion quality problem of image fusion,four parameters of pulse coupled neural networks （PCNN） model are set adopting fruit fly optimization algorithm （FOA）,and an image fusion method combining optimized PCNN and multi-scale transform is presented. Firstly,two registered origi- nal images are decomposed using four kinds of multi-scale transform separately. The selection principle of the low frequency component is the improved fusion method based on PCNN and FOA. The selection principle of the high frequency component is absolute maximum principle. Finally,the fused image is ob- tained by performing inverse multi-scale transform on the combined coefficients, subjectively, the new fu- sion rule not only preserves the details of original images well, but also improves the quality of fusion im- age ,objectively, fusion image of the new fusion rule is more superior in the objective evaluation indices, such as mutual information （MI）, edge preserving degree, entropy, mean structural similarity （MSSIM） and the standard deviation （SD）.

关 键 词：图像融合 多尺度变换 脉冲耦合神经网络(PCNN) 果蝇优化算法(FOA) 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]