摘要
为避免图像融合与超分辨率分步实现的不足,提出了基于卷积稀疏表示的融合与超分辨率重建联合实现方法。假设低分辨率与高分辨率图像之间具有相同的稀疏特征图,设计了一种高、低分辨率滤波器联合学习框架,实现对图像高低频成分的分离,并根据不同成分的形态特性设计了不同的融合规则:对于高频成分,根据稀疏特征图亮度信息和像素活跃性水平,设计了一种像素显著性度量方案来指导高频特征图的融合;对于低频成分,根据脉冲耦合神经网络能捕获邻域相似像素点火的特性,设计了低频成分融合方法。所提方法不需要将图像分割成重叠的块,避免块向量化的缺陷。实验结果表明,能有效提高图像融合的质量。
To avoid the defects of image fusion and super-resolution step-by-step implementation, a joint implementation method of fusion and super-resolution reconstruction based on convolutional sparse representation is proposed. This method assumes that there are the same sparse feature maps between low-resolution and high-resolution images, a joint learning framework of high-and low-resolution filters is designed to separate the high-frequency and low-frequency components. Furthermore, according to the morphological characteristics of different components, different fusion rules are designed. For high frequency component, according to the brightness information of sparse feature maps and the activity level of pixels, a scheme of measuring the saliency of pixels is designed to guide the fusion of high frequency feature maps. For low frequency component, a method of low-frequency component fusion is designed according to the characteristics of the impulse coupled neural network that can capture similar pixel ignition in the neighborhood. This method does not need to segment the image into overlapping blocks to avoid the defects of block vectorization. Experimental results show that this method can effectively improve the quality of image fusion.
作者
杨默远
李凡
谢明鸿
张亚飞
李华锋
YANG Moyuan;LI Fan;XIE Minhong;ZHANG Yafei;LI Huafeng(Faculty of Information Engineering and Automation,Kunming University of Science and Technology?Kunming 650500,China)
出处
《光学技术》
CAS
CSCD
北大核心
2020年第2期236-246,共11页
Optical Technique
基金
国家自然科学基金(61302041,61562053)。
关键词
图像融合
卷积稀疏表示
超分辨率
脉冲耦合神经网络
image fusion
convolutional sparse representation
super-resolution
pulse coupled neural network