基于鲁棒主成分分析的多聚焦图像融合

Multi-focus Image Fusion Based on Robust Principal Component Analysis

针对多聚焦图像融合提取聚焦区域不准确,融合图像的边缘存在伪影的问题,提出了一种基于鲁棒主成分分析的多聚焦图像融合方法。首先,用鲁棒主成分分析将源图像分解为低秩分量和稀疏分量。然后分析不同分量的特性,设计相应的融合规则。针对低秩分量,提出基于卷积神经网络的融合策略,用于提取源图像的纹理细节,构建权重图;对于稀疏分量,采用基于拉普拉斯能量和的方法生成决策图,为了使决策图的边缘与源图像保持一致,将多幅源图像的均值作为引导图像并用引导滤波器对决策图进行优化。最后,根据融合后的低秩分量和稀疏分量重构出最终的融合图像。实验结果表明,该方法提高了融合图像的对比度和边缘清晰度,在主观分析和客观评价指标上都优于其他七种方法。

In order to solve the problems of inaccurate focusing region extraction and artifacts in edge of image fusion,we propose a new image fusion algorithm based on robust principal component analysis.Firstly,the source images are decomposed into low-rank components and spare components by robust principal component analysis.Then the characteristics of different components is analyzed and the corresponding fusion rules are designed.For the low-rank components,a fusion rule based on convolutional neural network is proposed to extract the texture details of the source image and construct the weight map.For the sparse components,the Sum-Modified-Laplacian is used to generate the decision map.In order to keep the boundary of the decision map consistent with the source images,the mean value of multiple source images is used as the guided image in the guided filter to optimize the decision map.Finally,the final fused image is reconstructed from the fused low-rank component and the sparse component.The experiment shows that the proposed method improves the contrast and edge definition of the fused images,and is superior to other seven methods in both subjective and objective evaluation indicators.