基于深度学习理论的红外图像和可见光图像配准

Infrared image and visible image registration based on depth learning theory

为了获得理想图像,利用多源传感器的优点,提出了基于深度学习的红外图像和可见光图像配准方法。首先采用不同传感器采集红外图像和可见光图像,并对红外图像和可见光图像进行预处理,提取配准特征,然后采用深度学习理论根据特征进行红外图像和可见光图像配准,最后与其他方法进行了红外图像和可见光图像配准对比测试。结果表明,相对于对比方法,本方法的红外图像和可见光图像配准均方误差分别小0.14、0.15、0.04,峰值信噪比分别高8、4.5、11,归一化互信息分别高0.07、0.09、0.02,获得了更优的红外图像和可见光图像配准。

In order to obtain the ideal image, using the advantages of multi-source sensors, an infrared image and visible image registration method based on depth learning is proposed. First, the different sensors are collected to collect infrared and visible images, and the infrared and visible images are preprocessed to extract the registration features. Then, deep learning theory is used to match the infrared and visible images according to the characteristics. Finally, simultaneous interpreting and comparing the infrared and visible images with other methods. The results show that compared with the comparison method, the mean square error of infrared image and visible image registration of this method is 0.14, 0.15 and 0.04 respectively, the peak signal-to-noise ratio is 8, 4.5 and 11 respectively, and the normalized mutual information is 0.07, 0.09 and 0.02 respectively, so as to obtain better infrared image and visible image registration.