多方向独立小波耦合神经网络的图像融合算法

发布时间：2018-04-25 01:23

本文选题：图像融合 + 非分离小波多方向分析　；参考：《计算机工程与设计》2017年11期

【摘要】：针对当前图像融合算法因利用下采样分解导致信息冗余、融合图像对比度和清晰度不高等不足,提出基于非分离小波多方向分析NSWMDA(non-separable wavelet multi-directional analysis)与脉冲耦合神经网络PCNN(pulse coupled neural network)的图像融合算法。根据非分离小波变换和非下采样方向滤波器,设计多尺度NSWMDA机制,获取源图像的低频、高频子带;引入高斯-拉普拉斯算子,提取高频子带的高频信息,设计自适应PCNN融合准则,完成高频信息融合;基于高斯加权平均的低频融合规则,对低频子带完成融合;利用NSWMDA重构,输出融合图像。仿真结果表明,与当前图像融合算法相比,该算法的融合图像具有更优的交互信息、PSNR(peak signal to noise ratio),以及融合效率。
[Abstract]:In view of the shortcomings of current image fusion algorithms, such as information redundancy caused by downsampling decomposition, low contrast and clarity of fused images, etc. An image fusion algorithm based on unseparated wavelet multidirectional analysis (NSWMDA(non-separable wavelet multi-directional analysis) and impulsive coupled neural network (PCNN(pulse coupled neural network) is proposed. According to the unseparated wavelet transform and the non-downsampling directional filter, a multi-scale NSWMDA mechanism is designed to obtain the low frequency and high frequency subbands of the source image, and Gao Si Laplace operator is introduced to extract the high frequency information of the high frequency subband. Design adaptive PCNN fusion criterion to complete high frequency information fusion; based on Gao Si weighted average low frequency fusion rule to complete the fusion of low frequency subband; use NSWMDA reconstruction to output fusion image. The simulation results show that compared with the current image fusion algorithm, this algorithm has better interactive information signal to noise fusion efficiency.
【作者单位】：南宁学院信息工程学院;广西大学计算机与电子信息学院;南京理工大学计算机科学与工程学院;
【基金】：2015年广西省高校重点科研基金项目(KY2015Y13530) 广西自然科学基金项目(2014GXJSFAA019272)
【分类号】：TP183;TP391.41

【相似文献】