| 引用本文: | 翟浩,庄毅.利用拉普拉斯能量和CNN的多聚焦图像融合方法[J].哈尔滨工业大学学报,2020,52(5):137.DOI:10.11918/201909064 |
| ZHAI Hao,ZHUANG Yi.Multi-focus image fusion method using energy of Laplacian and convolutional neural network[J].Journal of Harbin Institute of Technology,2020,52(5):137.DOI:10.11918/201909064 |
|
| 摘要: |
| 多聚焦图像融合技术的目的是生成一幅全聚焦图像.所谓全聚焦图像,就是将不同源图像的清晰区域集成到一幅单一的图像中.传统的图像融合方法通常存在块伪影、人造边、晕轮效果、振铃效果以及对比度下降等问题.对此,本文提出了一种利用拉普拉斯能量和CNN的多聚焦图像融合方法.使用拉普拉斯能量算子可以有效的提取源图像的聚焦信息,而训练后的卷积神经网络模型从聚焦信息图中提取的聚焦特征可以有效的进行聚焦子块和离焦子块的区分.训练后的卷积神经网络模型不仅具有很好的提取活跃窗口相对聚焦度的能力,而且可以获得精确的分割边界.在经过多轮训练后,卷积神经网络模型可以很好的在源图像和分值图之间建立一种有效的映射,这对于生成一幅精准的聚焦图至关重要.采用二值分割和小区域滤波技术来对聚焦图进行进一步的修正,获得用于融合的最终决策图.最后,根据最终决策图提供的权值,对多幅源图像进行融合形成最终的融合图像.实验结果表明,无论从视觉效果还是从定量评价方面,提出的方法均优于目前已有的其它融合方法. |
| 关键词: 多聚焦图像融合 拉普拉斯能量 CNN 深度学习 聚焦度量 |
| DOI：10.11918/201909064 |
| 分类号:TP391.41 |
| 文献标识码:A |
| 基金项目:国家自然科学基金(61572253)； 航空科学基金(2016ZC52030)； 江苏省科技创新基金(KYLX16_0381) |
|
| Multi-focus image fusion method using energy of Laplacian and convolutional neural network |
|
ZHAI Hao,ZHUANG Yi
|
|
(College of Computer Science and Technology, Nanjing University of Aeronautics and Astronautics, Nanjing 211106, China)
|
| Abstract: |
| The aim of multi-focus image fusion technology is to produce an all-in-focus image in which the clear parts of different source images are integrated to a single image. Most of the existing fusion methods still suffer from the problems such as block artifacts, artificial edges, halo effects, ringing effects, and contrast reduction. To address these problems, a multi-focus image fusion method using the energy of Laplacian CNN is proposed in this paper. The focus information of source images can be extracted effectively by using Laplacian energy operator, and the focus feature extracted from focus information maps by the trained convolutional neural network model can effectively distinguish focused sub-blocks from defocused sub-blocks. The trained convolutional neural network model not only has a good ability to extract the relative focus degree of active windows, but also can obtain an accurate segmentation boundary. After multiple rounds of training, the convolutional neural network model can well establish an effective mapping between source images and a score map, which is essential to generate an accurate focus map. Then, the focus map is further modified using binary segmentation and small region filtering, and the final decision map for fusion is obtained. Finally, according to the weights provided by the final decision map, the final fusion image will be formed by fusing multiple source images. The experimental results show that the proposed method is superior to other existing fusion methods in terms of visual effects and quantitative evaluation. |
| Key words: multi-focus image fusion energy of Laplacian convolutional neural network deep learning focus measurement |
