Medical image fusion can effectively combine different images with different parts and complements each other, which is a hotspot in image processing. However, the medical image acquisition equipment usually obtains single image, so it is unable to use dataset to train convolutional neural network to extract stable image features. This paper proposes an approach using the Transfer Learning combined with the Limited Broyden, Fletcher, Goldfarb, and Shanno (L-BFGS) optimization algorithm to achieve single medical image fusion. The Transfer Learning which includes pre-trained model and parameters can be used as a feature extractor to extract or compress features in medical images. Firstly, different single medical images were input into the trained VGG16 Transfer Learning parameter model to extract image features, and then the maximum features between different images were obtained through feature comparison. Finally, L-BFGS optimization algorithm is used to approximate the initial image features to the maximum features, to achieve the fusion effect (the initial image can be any image of the same size and depth as the source image). The experimental results show that the algorithm is effective and very suitable for medical image fusion.

中文翻译：

---

使用迁移学习和 L-BFGS 优化算法的医学图像融合

医学图像融合可以有效地将不同图像的不同部分进行组合并相互补充，是图像处理中的一个热点。然而，医学图像采集设备通常获取单一图像，因此无法使用数据集训练卷积神经网络来提取稳定的图像特征。本文提出了一种使用迁移学习结合 Limited Broyden、Fletcher、Goldfarb 和 Shanno (L-BFGS) 优化算法来实现单张医学图像融合的方法。包括预训练模型和参数的迁移学习可用作特征提取器来提取或压缩医学图像中的特征。首先将不同的单张医学图像输入到训练好的VGG16迁移学习参数模型中提取图像特征，然后通过特征比较得到不同图像之间的最大特征。最后使用L-BFGS优化算法将初始图像特征逼近最大特征，达到融合效果（初始图像可以是与源图像大小和深度相同的任意图像）。实验结果表明，该算法是有效的，非常适用于医学图像融合。