非抽样双树复小波域的红外可见光图像融合  

Infrared and visible image fusion in undecimated dual-tree complex wavelet domain

作　　者：王旭辉[1] 周岩[1] 周苑[1] 

机构地区：[1]河南工程学院计算机学院,河南郑州451191

出　　处：《计算机工程与设计》2017年第3期729-734,共6页Computer Engineering and Design

基　　金：国家自然科学基金项目(61301232)

摘　　要：为提高红外与可见光图像的融合精度,借助非抽样双树复小波的平移不变性和良好的方向选择性,提出非负矩阵分解和区域方差能量的融合方法。通过对严格配准的红外与可见光图像进行非抽样双树复小波变换获取高低频信息,对低频子带系数采用基于块主元旋转的非负矩阵分解的融合方法,对高频子带系数选用结合区域方差能量的对比度进行融合。对融合后的系数采用非抽样双树复小波逆变换重构得到融合图像,对融合结果进行主观视觉和客观评价。对比实验结果表明,该算法具有较好的主观视觉效果,客观评价指标有明显提高,验证了提出算法的有效性。To improve the fusion accuracy of infrared visible image,an image fusion method based on non-negative matrix factorization and regional variance energy was proposed by means of the good directional selectivity and shift invariance of undecimated dual-tree complex wavelet.The high and low frequency information of images was obtained after the non-sampling dual-tree complex wavelet transformation for infrared and visible images with strict registration.The fusion principle of low frequency subband coefficients was based on the block principal pivoting method for nonnegative matrix factorization.For high frequency subband coefficients,the sum of contrast was adopted as the fusion rule combining with regional variance-energy.For low and high frequency information,the fusion image was reconstructed through undecimated dual-tree complex wavelet inverse transformation,and both subjective visual evaluation and objective performance assessments of the fusion results were implemented.The comparative experimental results show the proposed algorithm has good subjective visual effects and the objective evaluation index is obviously improved,which verifies the effectiveness of the algorithm.

关 键 词：图像融合 非抽样双树复小波变换 非负矩阵分解 块主元旋转法 加权区域能量 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]