摘要
针对传统Shearlet变换融合方法在图像的奇异处易产生伪吉布斯现象并且融合数据量较大的缺点,本文结合压缩感知(CS)理论提出一种基于改进Shearlet变换的红外与可见光图像融合方法。首先通过非下采样金字塔滤波器组对待融合红外与可见光图像分别进行分解,得到与原图像同等大小的高频信息图和低频信息图;对于低频信息图,采用能够恰当表征红外与可见光比例权值的局部区域信息熵进行融合;对于高频信息图,通过Shearlet变换中水平、垂直和对角剪切滤波器组进行滤波,继而将滤波后的系数分别采用简单且易物理实现的托普利茨矩阵进行观测并权值融合,然后通过分裂Bergman迭代获得融合后的高频Shearlet系数;最后经滤波器重构得到融合后的图像。实验结果表明该算法有效地解决了吉布斯现象问题,融合图像对比度较高,并相对于传统的融合方法减少了传输和融合的数据量,并且将融合时间缩短到50s以内,提高了融合效率。
Because the fusion method of traditional Shearlet transform may emerge pseudo Gibbs phenom- enon and has a large amount of data,this paper puts forward an infrared and visible image fusion method based on the improved Shearlet transform and the theory of compressed sensing. Shearlet transform can analyze image in more directions, and the inverse transform is only Shearlet filter. First of all, the pyra- mid filters are used to decomposite the images to get the high frequency and low frequency in formation graphs with the same size as the original images. For the low frequency information graph, the local area information entropy is used for fusion. For the high frequency information graph,the horizontal vertical and diagonal Shearlet filters are utilized and then the Toeplitz matrix is adopted to observe the filter co efficients, which is more simple and with easy physical implementation and then the split Bergman itera- tion is used to get the Shearlet coefficient for fusion. Finally,the fused image is got by filter reconstruc- tion. The experimental results show that the proposed method effectively avoids the pseudo-Gibbs phe- nomenon and outperforms the conventional fusion method on the amount of data transmission and fu- sion,and shortens the fusion time to less than 50,enhancing the efficiency of image fusion.
出处
《光电子．激光》
EI
CAS
CSCD
北大核心
2013年第2期384-390,共7页
Journal of Optoelectronics·Laser
基金
国家自然科学基金(61162016
60962004)资助项目
