基于Contourlet的图像融合方法  

Approach of Image Fusion Based on Contourlet Transform

作　　者：王建[1] 王必宁[2] 侯维刚[1] 

机构地区：[1]西安航空学院计算机学院,西安710077 [2]西安热工研究院有限公司,西安710032

出　　处：《微处理机》2016年第5期43-47,共5页Microprocessors

基　　金：陕西省光电测试与仪器技术重点实验室开放基金资助(ZSKJ201408)

摘　　要：针对两幅严格配准的同一场景左聚焦和右聚焦图像融合问题,设计了一种基于Contourlet变换的多聚焦图像融合方法。应用Contourlet变换对源图像进行多尺度、多方向分解,分别采用主成分分析法和区域结构相似度的融合规则得到融合图像的低通子带系数和带通方向子带系数,经过Contourlet逆变换得到融合图像。实验结果和定量分析结果表明,融合成为一幅清晰的单一聚焦图像,且将熵提高2.64%-3.15%,相关系数提高0.11%-0.57%,平均梯度提高4.44%-51.4%。Focusing on the problem of the fusion of the left - focus image and right - focus image from the same scene, a novel multi - focus image fusion approach based on contourlet transform is designed. Firstly, the contourlet transform is used to perform the multiscale and multidirection decomposition of the source images. Secondly, low frequency subband coefficients and bandpass directional subband coefficients of the fused image can be obtained by the fusion rules based on principal component analysis （PCA） and region structural similarity. Finally, the fused image is reconstructed by performing the inverse contourlet transform on the combined coefficients. The results of experiment and quantitative analysis demonstrate that the proposed algorithm can fuse into a clear single focus image, and the proposed algorithm is superior with entropy improvement from 2. 64% to 3. 15% , correlation coefficient enhancement from 0. 11% to 0. 57% and average gradient improvement from 4. 44% to 51.4% .

关 键 词：多聚焦图像 CONTOURLET变换 主成分分析 结构相似度 图像融合 融合规则 

分 类 号：TP39[自动化与计算机技术—计算机应用技术]