一种改进含噪多聚焦图像融合方法

发布时间：2018-01-05 00:32

本文关键词：一种改进含噪多聚焦图像融合方法　出处：《光电子·激光》2017年11期 　论文类型：期刊论文

【摘要】：针对目前多聚焦图像融合方法处理含噪图像缺乏有效性而导致融合效果较差的问题,提出一种引导滤波结合脉冲耦合神经网络(PCNN)的非下采样Shearlet变换(NSST)域内多聚焦图像融合方法。首先,分别对待融合多聚焦图像进行NSST获取其相应高频子带和低频子带系数;对高频子带系数,通过引导滤波结合改进简化PCNN模型设置融合规则;提取相位一致性、清晰度和亮度等底层视觉特性,指导低频子带系数融合权重;最后反NSST获取最终融合结果。实验结果表明,本文方法能够在噪声干扰情况下有效完成多聚焦融合,并且边缘和纹理信息保持较好,当20标准差噪声时互信息提升了近0.15具有有效性。
[Abstract]:In view of the problem that the current multi-focus image fusion method is not effective in processing noisy images, which leads to poor fusion effect. This paper presents a multi-focus image fusion method in the domain of non-downsampling Shearlet transform with guided filtering and pulse coupled neural network (PCNN). The fusion multi-focus images are treated with NSST to obtain the corresponding high-frequency subband and low-frequency subband coefficients. For the high frequency subband coefficient, the fusion rules of simplified PCNN model are improved by the combination of guided filtering and improved PCNN model. The low-level visual characteristics such as phase consistency, sharpness and brightness are extracted to guide the fusion weight of low frequency subband coefficients. Finally, the final fusion results are obtained by inverse NSST. Experimental results show that the proposed method can effectively achieve multi-focus fusion under noise interference, and the edge and texture information can be maintained well. When the noise is 20 standard deviation, the mutual information is improved by 0.15.
【作者单位】：重庆工商大学机械工程学院制造装备机构设计与控制重庆市重点实验室;中国农业科学院农业信息研究所;农业部规划设计研究院;
【基金】：国家自然科学基金(31501229) 重庆市基础科学与前沿技术研究(一般)项目(csct2015jcyjA40014) 重庆市教委基础与前沿研究计划(KJ1400628,KJ1500635)资助项目
【分类号】：TP183;TP391.41
【正文快照】： 1引言可见光成像系统通过调节光学镜头景深聚焦场景中的目标,无法同时清晰表达复杂场景的全部目标。多聚焦图像融合算法利用原始图像间的互补性,合并多幅原始图像的聚焦区域得到清晰*E-mail:149495263@qq.com的融合图像,能够更加准确和全面的描述图像中目标或场景,有效提高了

【相似文献】